openai_whisper-small.en_217MB/AudioEncoder.mlmodelc/model.mil
| 1 | program(1.0) |
| 2 | [buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3404.16.1"}, {"coremlc-version", "3404.23.1"}})] |
| 3 | { |
| 4 | func main<ios16>(tensor<fp16, [1, 80, 1, 3000]> melspectrogram_features) { |
| 5 | tensor<string, []> var_76_pad_type_0 = const()[name = tensor<string, []>("op_76_pad_type_0"), val = tensor<string, []>("custom")]; |
| 6 | tensor<int32, [4]> var_76_pad_0 = const()[name = tensor<string, []>("op_76_pad_0"), val = tensor<int32, [4]>([0, 0, 1, 1])]; |
| 7 | tensor<int32, [2]> var_76_strides_0 = const()[name = tensor<string, []>("op_76_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 8 | tensor<int32, [2]> var_76_dilations_0 = const()[name = tensor<string, []>("op_76_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 9 | tensor<int32, []> var_76_groups_0 = const()[name = tensor<string, []>("op_76_groups_0"), val = tensor<int32, []>(1)]; |
| 10 | tensor<fp16, [768, 80, 1, 3]> var_45_to_fp16 = const()[name = tensor<string, []>("op_45_to_fp16"), val = tensor<fp16, [768, 80, 1, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))]; |
| 11 | tensor<fp16, [768]> var_57_to_fp16 = const()[name = tensor<string, []>("op_57_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(368768)))]; |
| 12 | tensor<fp16, [1, 768, 1, 3000]> var_76_cast_fp16 = conv(bias = var_57_to_fp16, dilations = var_76_dilations_0, groups = var_76_groups_0, pad = var_76_pad_0, pad_type = var_76_pad_type_0, strides = var_76_strides_0, weight = var_45_to_fp16, x = melspectrogram_features)[name = tensor<string, []>("op_76_cast_fp16")]; |
| 13 | tensor<string, []> var_114_pad_type_0 = const()[name = tensor<string, []>("op_114_pad_type_0"), val = tensor<string, []>("custom")]; |
| 14 | tensor<int32, [4]> var_114_pad_0 = const()[name = tensor<string, []>("op_114_pad_0"), val = tensor<int32, [4]>([0, 0, 1, 1])]; |
| 15 | tensor<int32, [2]> var_114_strides_0 = const()[name = tensor<string, []>("op_114_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 16 | tensor<int32, [2]> var_114_dilations_0 = const()[name = tensor<string, []>("op_114_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 17 | tensor<int32, []> var_114_groups_0 = const()[name = tensor<string, []>("op_114_groups_0"), val = tensor<int32, []>(1)]; |
| 18 | tensor<fp16, [768, 80, 1, 3]> op_89_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [92160]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(370368))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(462592))), name = tensor<string, []>("op_89_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 80, 1, 3])]; |
| 19 | tensor<fp16, [768]> var_95_to_fp16 = const()[name = tensor<string, []>("op_95_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(462720)))]; |
| 20 | tensor<fp16, [1, 768, 1, 3000]> var_114_cast_fp16 = conv(bias = var_95_to_fp16, dilations = var_114_dilations_0, groups = var_114_groups_0, pad = var_114_pad_0, pad_type = var_114_pad_type_0, strides = var_114_strides_0, weight = op_89_to_fp16_palettized, x = melspectrogram_features)[name = tensor<string, []>("op_114_cast_fp16")]; |
| 21 | tensor<fp16, [1, 768, 1, 3000]> var_116_cast_fp16 = add(x = var_76_cast_fp16, y = var_114_cast_fp16)[name = tensor<string, []>("op_116_cast_fp16")]; |
| 22 | tensor<string, []> hidden_states_1_mode_0 = const()[name = tensor<string, []>("hidden_states_1_mode_0"), val = tensor<string, []>("EXACT")]; |
| 23 | tensor<fp16, [1, 768, 1, 3000]> hidden_states_1_cast_fp16 = gelu(mode = hidden_states_1_mode_0, x = var_116_cast_fp16)[name = tensor<string, []>("hidden_states_1_cast_fp16")]; |
| 24 | tensor<string, []> var_162_pad_type_0 = const()[name = tensor<string, []>("op_162_pad_type_0"), val = tensor<string, []>("custom")]; |
| 25 | tensor<int32, [4]> var_162_pad_0 = const()[name = tensor<string, []>("op_162_pad_0"), val = tensor<int32, [4]>([0, 0, 1, 1])]; |
| 26 | tensor<int32, [2]> var_162_strides_0 = const()[name = tensor<string, []>("op_162_strides_0"), val = tensor<int32, [2]>([2, 2])]; |
| 27 | tensor<int32, [2]> var_162_dilations_0 = const()[name = tensor<string, []>("op_162_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 28 | tensor<int32, []> var_162_groups_0 = const()[name = tensor<string, []>("op_162_groups_0"), val = tensor<int32, []>(1)]; |
| 29 | tensor<fp16, [768, 768, 1, 3]> var_131_to_fp16 = const()[name = tensor<string, []>("op_131_to_fp16"), val = tensor<fp16, [768, 768, 1, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(464320)))]; |
| 30 | tensor<fp16, [1, 768, 1, 1500]> var_162_cast_fp16 = conv(bias = var_57_to_fp16, dilations = var_162_dilations_0, groups = var_162_groups_0, pad = var_162_pad_0, pad_type = var_162_pad_type_0, strides = var_162_strides_0, weight = var_131_to_fp16, x = hidden_states_1_cast_fp16)[name = tensor<string, []>("op_162_cast_fp16")]; |
| 31 | tensor<string, []> var_200_pad_type_0 = const()[name = tensor<string, []>("op_200_pad_type_0"), val = tensor<string, []>("custom")]; |
| 32 | tensor<int32, [4]> var_200_pad_0 = const()[name = tensor<string, []>("op_200_pad_0"), val = tensor<int32, [4]>([0, 0, 1, 1])]; |
| 33 | tensor<int32, [2]> var_200_strides_0 = const()[name = tensor<string, []>("op_200_strides_0"), val = tensor<int32, [2]>([2, 2])]; |
| 34 | tensor<int32, [2]> var_200_dilations_0 = const()[name = tensor<string, []>("op_200_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 35 | tensor<int32, []> var_200_groups_0 = const()[name = tensor<string, []>("op_200_groups_0"), val = tensor<int32, []>(1)]; |
| 36 | tensor<fp16, [768, 768, 1, 3]> op_175_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [884736]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4003328))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4888128))), name = tensor<string, []>("op_175_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 3])]; |
| 37 | tensor<fp16, [768]> var_181_to_fp16 = const()[name = tensor<string, []>("op_181_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4888256)))]; |
| 38 | tensor<fp16, [1, 768, 1, 1500]> var_200_cast_fp16 = conv(bias = var_181_to_fp16, dilations = var_200_dilations_0, groups = var_200_groups_0, pad = var_200_pad_0, pad_type = var_200_pad_type_0, strides = var_200_strides_0, weight = op_175_to_fp16_palettized, x = hidden_states_1_cast_fp16)[name = tensor<string, []>("op_200_cast_fp16")]; |
| 39 | tensor<fp16, [1, 768, 1, 1500]> var_202_cast_fp16 = add(x = var_162_cast_fp16, y = var_200_cast_fp16)[name = tensor<string, []>("op_202_cast_fp16")]; |
| 40 | tensor<string, []> hidden_states_3_mode_0 = const()[name = tensor<string, []>("hidden_states_3_mode_0"), val = tensor<string, []>("EXACT")]; |
| 41 | tensor<fp16, [1, 768, 1, 1500]> hidden_states_3_cast_fp16 = gelu(mode = hidden_states_3_mode_0, x = var_202_cast_fp16)[name = tensor<string, []>("hidden_states_3_cast_fp16")]; |
| 42 | tensor<fp16, [1, 768, 1, 1500]> var_222_to_fp16 = const()[name = tensor<string, []>("op_222_to_fp16"), val = tensor<fp16, [1, 768, 1, 1500]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4889856)))]; |
| 43 | tensor<fp16, [1, 768, 1, 1500]> inputs_1_cast_fp16 = add(x = hidden_states_3_cast_fp16, y = var_222_to_fp16)[name = tensor<string, []>("inputs_1_cast_fp16")]; |
| 44 | tensor<int32, []> var_232 = const()[name = tensor<string, []>("op_232"), val = tensor<int32, []>(3)]; |
| 45 | tensor<int32, [1]> out_1_axes_0 = const()[name = tensor<string, []>("out_1_axes_0"), val = tensor<int32, [1]>([1])]; |
| 46 | tensor<fp16, []> var_254_to_fp16 = const()[name = tensor<string, []>("op_254_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 47 | tensor<fp16, [1, 768, 1, 1500]> out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_254_to_fp16, x = inputs_1_cast_fp16)[name = tensor<string, []>("out_1_cast_fp16")]; |
| 48 | tensor<fp16, [768]> obj_1_variance_0_to_fp16 = const()[name = tensor<string, []>("obj_1_variance_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7193920)))]; |
| 49 | tensor<fp16, [768]> obj_1_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_1_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7195520)))]; |
| 50 | tensor<fp16, [768]> obj_1_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_1_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7197120)))]; |
| 51 | tensor<fp16, []> obj_1_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_1_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 52 | tensor<fp16, [1, 768, 1, 1500]> obj_1_cast_fp16 = batch_norm(beta = obj_1_beta_0_to_fp16, epsilon = obj_1_epsilon_0_to_fp16, gamma = obj_1_gamma_0_to_fp16, mean = var_57_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_1_cast_fp16)[name = tensor<string, []>("obj_1_cast_fp16")]; |
| 53 | tensor<string, []> var_276_pad_type_0 = const()[name = tensor<string, []>("op_276_pad_type_0"), val = tensor<string, []>("valid")]; |
| 54 | tensor<int32, [2]> var_276_strides_0 = const()[name = tensor<string, []>("op_276_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 55 | tensor<int32, [4]> var_276_pad_0 = const()[name = tensor<string, []>("op_276_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 56 | tensor<int32, [2]> var_276_dilations_0 = const()[name = tensor<string, []>("op_276_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 57 | tensor<int32, []> var_276_groups_0 = const()[name = tensor<string, []>("op_276_groups_0"), val = tensor<int32, []>(1)]; |
| 58 | tensor<fp16, [768, 768, 1, 1]> layers_0_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7198720))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7493696))), name = tensor<string, []>("layers_0_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 59 | tensor<fp16, [768]> layers_0_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7493824)))]; |
| 60 | tensor<fp16, [1, 768, 1, 1500]> var_276_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_276_dilations_0, groups = var_276_groups_0, pad = var_276_pad_0, pad_type = var_276_pad_type_0, strides = var_276_strides_0, weight = layers_0_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_1_cast_fp16)[name = tensor<string, []>("op_276_cast_fp16")]; |
| 61 | tensor<string, []> var_282_pad_type_0 = const()[name = tensor<string, []>("op_282_pad_type_0"), val = tensor<string, []>("valid")]; |
| 62 | tensor<int32, [2]> var_282_strides_0 = const()[name = tensor<string, []>("op_282_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 63 | tensor<int32, [4]> var_282_pad_0 = const()[name = tensor<string, []>("op_282_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 64 | tensor<int32, [2]> var_282_dilations_0 = const()[name = tensor<string, []>("op_282_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 65 | tensor<int32, []> var_282_groups_0 = const()[name = tensor<string, []>("op_282_groups_0"), val = tensor<int32, []>(1)]; |
| 66 | tensor<fp16, [768, 768, 1, 1]> layers_0_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7518208))), name = tensor<string, []>("layers_0_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [11338]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7495424))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 67 | tensor<fp16, [1, 768, 1, 1500]> var_282_cast_fp16 = conv(dilations = var_282_dilations_0, groups = var_282_groups_0, pad = var_282_pad_0, pad_type = var_282_pad_type_0, strides = var_282_strides_0, weight = layers_0_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_1_cast_fp16)[name = tensor<string, []>("op_282_cast_fp16")]; |
| 68 | tensor<fp16, [1, 768, 1, 1500]> query_1_cast_fp16 = add(x = var_276_cast_fp16, y = var_282_cast_fp16)[name = tensor<string, []>("query_1_cast_fp16")]; |
| 69 | tensor<string, []> var_291_pad_type_0 = const()[name = tensor<string, []>("op_291_pad_type_0"), val = tensor<string, []>("valid")]; |
| 70 | tensor<int32, [2]> var_291_strides_0 = const()[name = tensor<string, []>("op_291_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 71 | tensor<int32, [4]> var_291_pad_0 = const()[name = tensor<string, []>("op_291_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 72 | tensor<int32, [2]> var_291_dilations_0 = const()[name = tensor<string, []>("op_291_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 73 | tensor<int32, []> var_291_groups_0 = const()[name = tensor<string, []>("op_291_groups_0"), val = tensor<int32, []>(1)]; |
| 74 | tensor<fp16, [768, 768, 1, 1]> layers_0_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7592000))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7886976))), name = tensor<string, []>("layers_0_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 75 | tensor<fp16, [1, 768, 1, 1500]> var_291_cast_fp16 = conv(dilations = var_291_dilations_0, groups = var_291_groups_0, pad = var_291_pad_0, pad_type = var_291_pad_type_0, strides = var_291_strides_0, weight = layers_0_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_1_cast_fp16)[name = tensor<string, []>("op_291_cast_fp16")]; |
| 76 | tensor<string, []> var_297_pad_type_0 = const()[name = tensor<string, []>("op_297_pad_type_0"), val = tensor<string, []>("valid")]; |
| 77 | tensor<int32, [2]> var_297_strides_0 = const()[name = tensor<string, []>("op_297_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 78 | tensor<int32, [4]> var_297_pad_0 = const()[name = tensor<string, []>("op_297_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 79 | tensor<int32, [2]> var_297_dilations_0 = const()[name = tensor<string, []>("op_297_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 80 | tensor<int32, []> var_297_groups_0 = const()[name = tensor<string, []>("op_297_groups_0"), val = tensor<int32, []>(1)]; |
| 81 | tensor<fp16, [768, 768, 1, 1]> layers_0_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7908352))), name = tensor<string, []>("layers_0_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [10583]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7887104))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 82 | tensor<fp16, [1, 768, 1, 1500]> var_297_cast_fp16 = conv(dilations = var_297_dilations_0, groups = var_297_groups_0, pad = var_297_pad_0, pad_type = var_297_pad_type_0, strides = var_297_strides_0, weight = layers_0_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_1_cast_fp16)[name = tensor<string, []>("op_297_cast_fp16")]; |
| 83 | tensor<fp16, [1, 768, 1, 1500]> key_1_cast_fp16 = add(x = var_291_cast_fp16, y = var_297_cast_fp16)[name = tensor<string, []>("key_1_cast_fp16")]; |
| 84 | tensor<string, []> var_307_pad_type_0 = const()[name = tensor<string, []>("op_307_pad_type_0"), val = tensor<string, []>("valid")]; |
| 85 | tensor<int32, [2]> var_307_strides_0 = const()[name = tensor<string, []>("op_307_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 86 | tensor<int32, [4]> var_307_pad_0 = const()[name = tensor<string, []>("op_307_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 87 | tensor<int32, [2]> var_307_dilations_0 = const()[name = tensor<string, []>("op_307_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 88 | tensor<int32, []> var_307_groups_0 = const()[name = tensor<string, []>("op_307_groups_0"), val = tensor<int32, []>(1)]; |
| 89 | tensor<fp16, [768, 768, 1, 1]> layers_0_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7982144))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8277120))), name = tensor<string, []>("layers_0_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 90 | tensor<fp16, [768]> layers_0_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8277248)))]; |
| 91 | tensor<fp16, [1, 768, 1, 1500]> var_307_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_307_dilations_0, groups = var_307_groups_0, pad = var_307_pad_0, pad_type = var_307_pad_type_0, strides = var_307_strides_0, weight = layers_0_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_1_cast_fp16)[name = tensor<string, []>("op_307_cast_fp16")]; |
| 92 | tensor<string, []> var_313_pad_type_0 = const()[name = tensor<string, []>("op_313_pad_type_0"), val = tensor<string, []>("valid")]; |
| 93 | tensor<int32, [2]> var_313_strides_0 = const()[name = tensor<string, []>("op_313_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 94 | tensor<int32, [4]> var_313_pad_0 = const()[name = tensor<string, []>("op_313_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 95 | tensor<int32, [2]> var_313_dilations_0 = const()[name = tensor<string, []>("op_313_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 96 | tensor<int32, []> var_313_groups_0 = const()[name = tensor<string, []>("op_313_groups_0"), val = tensor<int32, []>(1)]; |
| 97 | tensor<fp16, [768, 768, 1, 1]> layers_0_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8302400))), name = tensor<string, []>("layers_0_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [11740]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8278848))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 98 | tensor<fp16, [1, 768, 1, 1500]> var_313_cast_fp16 = conv(dilations = var_313_dilations_0, groups = var_313_groups_0, pad = var_313_pad_0, pad_type = var_313_pad_type_0, strides = var_313_strides_0, weight = layers_0_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_1_cast_fp16)[name = tensor<string, []>("op_313_cast_fp16")]; |
| 99 | tensor<fp16, [1, 768, 1, 1500]> value_1_cast_fp16 = add(x = var_307_cast_fp16, y = var_313_cast_fp16)[name = tensor<string, []>("value_1_cast_fp16")]; |
| 100 | tensor<int32, [4]> var_317 = const()[name = tensor<string, []>("op_317"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 101 | tensor<fp16, [1, 12, 64, 1500]> mh_q_1_cast_fp16 = reshape(shape = var_317, x = query_1_cast_fp16)[name = tensor<string, []>("mh_q_1_cast_fp16")]; |
| 102 | tensor<fp16, []> var_319_to_fp16 = const()[name = tensor<string, []>("op_319_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| 103 | tensor<fp16, [1, 12, 64, 1500]> var_320_cast_fp16 = mul(x = mh_q_1_cast_fp16, y = var_319_to_fp16)[name = tensor<string, []>("op_320_cast_fp16")]; |
| 104 | tensor<int32, [4]> var_323 = const()[name = tensor<string, []>("op_323"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 105 | tensor<fp16, [1, 12, 64, 1500]> var_324_cast_fp16 = reshape(shape = var_323, x = key_1_cast_fp16)[name = tensor<string, []>("op_324_cast_fp16")]; |
| 106 | tensor<bool, []> mh_w_1_transpose_x_0 = const()[name = tensor<string, []>("mh_w_1_transpose_x_0"), val = tensor<bool, []>(true)]; |
| 107 | tensor<bool, []> mh_w_1_transpose_y_0 = const()[name = tensor<string, []>("mh_w_1_transpose_y_0"), val = tensor<bool, []>(false)]; |
| 108 | tensor<fp16, [1, 12, 1500, 1500]> mh_w_1_cast_fp16 = matmul(transpose_x = mh_w_1_transpose_x_0, transpose_y = mh_w_1_transpose_y_0, x = var_320_cast_fp16, y = var_324_cast_fp16)[name = tensor<string, []>("mh_w_1_cast_fp16")]; |
| 109 | tensor<fp16, [1, 12, 1500, 1500]> var_327_cast_fp16 = softmax(axis = var_232, x = mh_w_1_cast_fp16)[name = tensor<string, []>("op_327_cast_fp16")]; |
| 110 | tensor<int32, [4]> var_328 = const()[name = tensor<string, []>("op_328"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 111 | tensor<fp16, [1, 12, 64, 1500]> var_329_cast_fp16 = reshape(shape = var_328, x = value_1_cast_fp16)[name = tensor<string, []>("op_329_cast_fp16")]; |
| 112 | tensor<bool, []> attn_1_transpose_x_0 = const()[name = tensor<string, []>("attn_1_transpose_x_0"), val = tensor<bool, []>(false)]; |
| 113 | tensor<bool, []> attn_1_transpose_y_0 = const()[name = tensor<string, []>("attn_1_transpose_y_0"), val = tensor<bool, []>(true)]; |
| 114 | tensor<fp16, [1, 12, 64, 1500]> attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_329_cast_fp16, y = var_327_cast_fp16)[name = tensor<string, []>("attn_1_cast_fp16")]; |
| 115 | tensor<int32, [4]> var_332 = const()[name = tensor<string, []>("op_332"), val = tensor<int32, [4]>([1, 768, 1, 1500])]; |
| 116 | tensor<fp16, [1, 768, 1, 1500]> input_1_cast_fp16 = reshape(shape = var_332, x = attn_1_cast_fp16)[name = tensor<string, []>("input_1_cast_fp16")]; |
| 117 | tensor<string, []> var_342_pad_type_0 = const()[name = tensor<string, []>("op_342_pad_type_0"), val = tensor<string, []>("valid")]; |
| 118 | tensor<int32, [2]> var_342_strides_0 = const()[name = tensor<string, []>("op_342_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 119 | tensor<int32, [4]> var_342_pad_0 = const()[name = tensor<string, []>("op_342_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 120 | tensor<int32, [2]> var_342_dilations_0 = const()[name = tensor<string, []>("op_342_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 121 | tensor<int32, []> var_342_groups_0 = const()[name = tensor<string, []>("op_342_groups_0"), val = tensor<int32, []>(1)]; |
| 122 | tensor<fp16, [768, 768, 1, 1]> layers_0_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8376192))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8671168))), name = tensor<string, []>("layers_0_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 123 | tensor<fp16, [768]> layers_0_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8671296)))]; |
| 124 | tensor<fp16, [1, 768, 1, 1500]> var_342_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_342_dilations_0, groups = var_342_groups_0, pad = var_342_pad_0, pad_type = var_342_pad_type_0, strides = var_342_strides_0, weight = layers_0_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_1_cast_fp16)[name = tensor<string, []>("op_342_cast_fp16")]; |
| 125 | tensor<string, []> var_348_pad_type_0 = const()[name = tensor<string, []>("op_348_pad_type_0"), val = tensor<string, []>("valid")]; |
| 126 | tensor<int32, [2]> var_348_strides_0 = const()[name = tensor<string, []>("op_348_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 127 | tensor<int32, [4]> var_348_pad_0 = const()[name = tensor<string, []>("op_348_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 128 | tensor<int32, [2]> var_348_dilations_0 = const()[name = tensor<string, []>("op_348_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 129 | tensor<int32, []> var_348_groups_0 = const()[name = tensor<string, []>("op_348_groups_0"), val = tensor<int32, []>(1)]; |
| 130 | tensor<fp16, [768, 768, 1, 1]> layers_0_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8692480))), name = tensor<string, []>("layers_0_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [9758]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8672896))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 131 | tensor<fp16, [1, 768, 1, 1500]> var_348_cast_fp16 = conv(dilations = var_348_dilations_0, groups = var_348_groups_0, pad = var_348_pad_0, pad_type = var_348_pad_type_0, strides = var_348_strides_0, weight = layers_0_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_1_cast_fp16)[name = tensor<string, []>("op_348_cast_fp16")]; |
| 132 | tensor<fp16, [1, 768, 1, 1500]> obj_3_cast_fp16 = add(x = var_342_cast_fp16, y = var_348_cast_fp16)[name = tensor<string, []>("obj_3_cast_fp16")]; |
| 133 | tensor<fp16, [1, 768, 1, 1500]> inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_3_cast_fp16)[name = tensor<string, []>("inputs_3_cast_fp16")]; |
| 134 | tensor<int32, [1]> out_3_axes_0 = const()[name = tensor<string, []>("out_3_axes_0"), val = tensor<int32, [1]>([1])]; |
| 135 | tensor<fp16, []> var_359_to_fp16 = const()[name = tensor<string, []>("op_359_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 136 | tensor<fp16, [1, 768, 1, 1500]> out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_359_to_fp16, x = inputs_3_cast_fp16)[name = tensor<string, []>("out_3_cast_fp16")]; |
| 137 | tensor<fp16, [768]> input_3_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_3_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8766272)))]; |
| 138 | tensor<fp16, [768]> input_3_beta_0_to_fp16 = const()[name = tensor<string, []>("input_3_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8767872)))]; |
| 139 | tensor<fp16, []> input_3_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_3_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 140 | tensor<fp16, [1, 768, 1, 1500]> input_3_cast_fp16 = batch_norm(beta = input_3_beta_0_to_fp16, epsilon = input_3_epsilon_0_to_fp16, gamma = input_3_gamma_0_to_fp16, mean = var_57_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_3_cast_fp16)[name = tensor<string, []>("input_3_cast_fp16")]; |
| 141 | tensor<string, []> var_377_pad_type_0 = const()[name = tensor<string, []>("op_377_pad_type_0"), val = tensor<string, []>("valid")]; |
| 142 | tensor<int32, [2]> var_377_strides_0 = const()[name = tensor<string, []>("op_377_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 143 | tensor<int32, [4]> var_377_pad_0 = const()[name = tensor<string, []>("op_377_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 144 | tensor<int32, [2]> var_377_dilations_0 = const()[name = tensor<string, []>("op_377_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 145 | tensor<int32, []> var_377_groups_0 = const()[name = tensor<string, []>("op_377_groups_0"), val = tensor<int32, []>(1)]; |
| 146 | tensor<fp16, [3072, 768, 1, 1]> layers_0_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1179648]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8769472))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9949184))), name = tensor<string, []>("layers_0_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([3072, 768, 1, 1])]; |
| 147 | tensor<fp16, [3072]> layers_0_fc1_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_fc1_inlier_module_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9949312)))]; |
| 148 | tensor<fp16, [1, 3072, 1, 1500]> var_377_cast_fp16 = conv(bias = layers_0_fc1_inlier_module_bias_to_fp16, dilations = var_377_dilations_0, groups = var_377_groups_0, pad = var_377_pad_0, pad_type = var_377_pad_type_0, strides = var_377_strides_0, weight = layers_0_fc1_inlier_module_weight_to_fp16_palettized, x = input_3_cast_fp16)[name = tensor<string, []>("op_377_cast_fp16")]; |
| 149 | tensor<string, []> var_383_pad_type_0 = const()[name = tensor<string, []>("op_383_pad_type_0"), val = tensor<string, []>("valid")]; |
| 150 | tensor<int32, [2]> var_383_strides_0 = const()[name = tensor<string, []>("op_383_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 151 | tensor<int32, [4]> var_383_pad_0 = const()[name = tensor<string, []>("op_383_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 152 | tensor<int32, [2]> var_383_dilations_0 = const()[name = tensor<string, []>("op_383_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 153 | tensor<int32, []> var_383_groups_0 = const()[name = tensor<string, []>("op_383_groups_0"), val = tensor<int32, []>(1)]; |
| 154 | tensor<fp16, [3072, 768, 1, 1]> layers_0_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10044736))), name = tensor<string, []>("layers_0_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [44545]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9955520))), shape = tensor<uint32, [4]>([3072, 768, 1, 1])]; |
| 155 | tensor<fp16, [1, 3072, 1, 1500]> var_383_cast_fp16 = conv(dilations = var_383_dilations_0, groups = var_383_groups_0, pad = var_383_pad_0, pad_type = var_383_pad_type_0, strides = var_383_strides_0, weight = layers_0_fc1_outlier_module_weight_to_fp16_sparsified, x = input_3_cast_fp16)[name = tensor<string, []>("op_383_cast_fp16")]; |
| 156 | tensor<fp16, [1, 3072, 1, 1500]> input_5_cast_fp16 = add(x = var_377_cast_fp16, y = var_383_cast_fp16)[name = tensor<string, []>("input_5_cast_fp16")]; |
| 157 | tensor<string, []> input_7_mode_0 = const()[name = tensor<string, []>("input_7_mode_0"), val = tensor<string, []>("EXACT")]; |
| 158 | tensor<fp16, [1, 3072, 1, 1500]> input_7_cast_fp16 = gelu(mode = input_7_mode_0, x = input_5_cast_fp16)[name = tensor<string, []>("input_7_cast_fp16")]; |
| 159 | tensor<string, []> var_394_pad_type_0 = const()[name = tensor<string, []>("op_394_pad_type_0"), val = tensor<string, []>("valid")]; |
| 160 | tensor<int32, [2]> var_394_strides_0 = const()[name = tensor<string, []>("op_394_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 161 | tensor<int32, [4]> var_394_pad_0 = const()[name = tensor<string, []>("op_394_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 162 | tensor<int32, [2]> var_394_dilations_0 = const()[name = tensor<string, []>("op_394_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 163 | tensor<int32, []> var_394_groups_0 = const()[name = tensor<string, []>("op_394_groups_0"), val = tensor<int32, []>(1)]; |
| 164 | tensor<fp16, [768, 3072, 1, 1]> layers_0_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1179648]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10339712))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11519424))), name = tensor<string, []>("layers_0_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 3072, 1, 1])]; |
| 165 | tensor<fp16, [768]> layers_0_fc2_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_fc2_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11519552)))]; |
| 166 | tensor<fp16, [1, 768, 1, 1500]> var_394_cast_fp16 = conv(bias = layers_0_fc2_inlier_module_bias_to_fp16, dilations = var_394_dilations_0, groups = var_394_groups_0, pad = var_394_pad_0, pad_type = var_394_pad_type_0, strides = var_394_strides_0, weight = layers_0_fc2_inlier_module_weight_to_fp16_palettized, x = input_7_cast_fp16)[name = tensor<string, []>("op_394_cast_fp16")]; |
| 167 | tensor<string, []> var_400_pad_type_0 = const()[name = tensor<string, []>("op_400_pad_type_0"), val = tensor<string, []>("valid")]; |
| 168 | tensor<int32, [2]> var_400_strides_0 = const()[name = tensor<string, []>("op_400_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 169 | tensor<int32, [4]> var_400_pad_0 = const()[name = tensor<string, []>("op_400_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 170 | tensor<int32, [2]> var_400_dilations_0 = const()[name = tensor<string, []>("op_400_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 171 | tensor<int32, []> var_400_groups_0 = const()[name = tensor<string, []>("op_400_groups_0"), val = tensor<int32, []>(1)]; |
| 172 | tensor<fp16, [768, 3072, 1, 1]> layers_0_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11594560))), name = tensor<string, []>("layers_0_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [36651]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11521152))), shape = tensor<uint32, [4]>([768, 3072, 1, 1])]; |
| 173 | tensor<fp16, [1, 768, 1, 1500]> var_400_cast_fp16 = conv(dilations = var_400_dilations_0, groups = var_400_groups_0, pad = var_400_pad_0, pad_type = var_400_pad_type_0, strides = var_400_strides_0, weight = layers_0_fc2_outlier_module_weight_to_fp16_sparsified, x = input_7_cast_fp16)[name = tensor<string, []>("op_400_cast_fp16")]; |
| 174 | tensor<fp16, [1, 768, 1, 1500]> hidden_states_5_cast_fp16 = add(x = var_394_cast_fp16, y = var_400_cast_fp16)[name = tensor<string, []>("hidden_states_5_cast_fp16")]; |
| 175 | tensor<fp16, [1, 768, 1, 1500]> inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = hidden_states_5_cast_fp16)[name = tensor<string, []>("inputs_5_cast_fp16")]; |
| 176 | tensor<int32, []> var_406 = const()[name = tensor<string, []>("op_406"), val = tensor<int32, []>(3)]; |
| 177 | tensor<int32, [1]> out_5_axes_0 = const()[name = tensor<string, []>("out_5_axes_0"), val = tensor<int32, [1]>([1])]; |
| 178 | tensor<fp16, []> var_428_to_fp16 = const()[name = tensor<string, []>("op_428_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 179 | tensor<fp16, [1, 768, 1, 1500]> out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_428_to_fp16, x = inputs_5_cast_fp16)[name = tensor<string, []>("out_5_cast_fp16")]; |
| 180 | tensor<fp16, [768]> obj_5_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_5_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11889536)))]; |
| 181 | tensor<fp16, [768]> obj_5_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_5_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11891136)))]; |
| 182 | tensor<fp16, []> obj_5_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_5_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 183 | tensor<fp16, [1, 768, 1, 1500]> obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = var_57_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_5_cast_fp16)[name = tensor<string, []>("obj_5_cast_fp16")]; |
| 184 | tensor<string, []> var_450_pad_type_0 = const()[name = tensor<string, []>("op_450_pad_type_0"), val = tensor<string, []>("valid")]; |
| 185 | tensor<int32, [2]> var_450_strides_0 = const()[name = tensor<string, []>("op_450_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 186 | tensor<int32, [4]> var_450_pad_0 = const()[name = tensor<string, []>("op_450_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 187 | tensor<int32, [2]> var_450_dilations_0 = const()[name = tensor<string, []>("op_450_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 188 | tensor<int32, []> var_450_groups_0 = const()[name = tensor<string, []>("op_450_groups_0"), val = tensor<int32, []>(1)]; |
| 189 | tensor<fp16, [768, 768, 1, 1]> layers_1_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11892736))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12187712))), name = tensor<string, []>("layers_1_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 190 | tensor<fp16, [768]> layers_1_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12187840)))]; |
| 191 | tensor<fp16, [1, 768, 1, 1500]> var_450_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_450_dilations_0, groups = var_450_groups_0, pad = var_450_pad_0, pad_type = var_450_pad_type_0, strides = var_450_strides_0, weight = layers_1_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_5_cast_fp16)[name = tensor<string, []>("op_450_cast_fp16")]; |
| 192 | tensor<string, []> var_456_pad_type_0 = const()[name = tensor<string, []>("op_456_pad_type_0"), val = tensor<string, []>("valid")]; |
| 193 | tensor<int32, [2]> var_456_strides_0 = const()[name = tensor<string, []>("op_456_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 194 | tensor<int32, [4]> var_456_pad_0 = const()[name = tensor<string, []>("op_456_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 195 | tensor<int32, [2]> var_456_dilations_0 = const()[name = tensor<string, []>("op_456_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 196 | tensor<int32, []> var_456_groups_0 = const()[name = tensor<string, []>("op_456_groups_0"), val = tensor<int32, []>(1)]; |
| 197 | tensor<fp16, [768, 768, 1, 1]> layers_1_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12206656))), name = tensor<string, []>("layers_1_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [8573]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12189440))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 198 | tensor<fp16, [1, 768, 1, 1500]> var_456_cast_fp16 = conv(dilations = var_456_dilations_0, groups = var_456_groups_0, pad = var_456_pad_0, pad_type = var_456_pad_type_0, strides = var_456_strides_0, weight = layers_1_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_5_cast_fp16)[name = tensor<string, []>("op_456_cast_fp16")]; |
| 199 | tensor<fp16, [1, 768, 1, 1500]> query_3_cast_fp16 = add(x = var_450_cast_fp16, y = var_456_cast_fp16)[name = tensor<string, []>("query_3_cast_fp16")]; |
| 200 | tensor<string, []> var_465_pad_type_0 = const()[name = tensor<string, []>("op_465_pad_type_0"), val = tensor<string, []>("valid")]; |
| 201 | tensor<int32, [2]> var_465_strides_0 = const()[name = tensor<string, []>("op_465_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 202 | tensor<int32, [4]> var_465_pad_0 = const()[name = tensor<string, []>("op_465_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 203 | tensor<int32, [2]> var_465_dilations_0 = const()[name = tensor<string, []>("op_465_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 204 | tensor<int32, []> var_465_groups_0 = const()[name = tensor<string, []>("op_465_groups_0"), val = tensor<int32, []>(1)]; |
| 205 | tensor<fp16, [768, 768, 1, 1]> layers_1_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12280448))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12575424))), name = tensor<string, []>("layers_1_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 206 | tensor<fp16, [1, 768, 1, 1500]> var_465_cast_fp16 = conv(dilations = var_465_dilations_0, groups = var_465_groups_0, pad = var_465_pad_0, pad_type = var_465_pad_type_0, strides = var_465_strides_0, weight = layers_1_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_5_cast_fp16)[name = tensor<string, []>("op_465_cast_fp16")]; |
| 207 | tensor<string, []> var_471_pad_type_0 = const()[name = tensor<string, []>("op_471_pad_type_0"), val = tensor<string, []>("valid")]; |
| 208 | tensor<int32, [2]> var_471_strides_0 = const()[name = tensor<string, []>("op_471_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 209 | tensor<int32, [4]> var_471_pad_0 = const()[name = tensor<string, []>("op_471_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 210 | tensor<int32, [2]> var_471_dilations_0 = const()[name = tensor<string, []>("op_471_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 211 | tensor<int32, []> var_471_groups_0 = const()[name = tensor<string, []>("op_471_groups_0"), val = tensor<int32, []>(1)]; |
| 212 | tensor<fp16, [768, 768, 1, 1]> layers_1_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12591360))), name = tensor<string, []>("layers_1_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [7864]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12575552))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 213 | tensor<fp16, [1, 768, 1, 1500]> var_471_cast_fp16 = conv(dilations = var_471_dilations_0, groups = var_471_groups_0, pad = var_471_pad_0, pad_type = var_471_pad_type_0, strides = var_471_strides_0, weight = layers_1_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_5_cast_fp16)[name = tensor<string, []>("op_471_cast_fp16")]; |
| 214 | tensor<fp16, [1, 768, 1, 1500]> key_3_cast_fp16 = add(x = var_465_cast_fp16, y = var_471_cast_fp16)[name = tensor<string, []>("key_3_cast_fp16")]; |
| 215 | tensor<string, []> var_481_pad_type_0 = const()[name = tensor<string, []>("op_481_pad_type_0"), val = tensor<string, []>("valid")]; |
| 216 | tensor<int32, [2]> var_481_strides_0 = const()[name = tensor<string, []>("op_481_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 217 | tensor<int32, [4]> var_481_pad_0 = const()[name = tensor<string, []>("op_481_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 218 | tensor<int32, [2]> var_481_dilations_0 = const()[name = tensor<string, []>("op_481_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 219 | tensor<int32, []> var_481_groups_0 = const()[name = tensor<string, []>("op_481_groups_0"), val = tensor<int32, []>(1)]; |
| 220 | tensor<fp16, [768, 768, 1, 1]> layers_1_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12665152))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12960128))), name = tensor<string, []>("layers_1_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 221 | tensor<fp16, [768]> layers_1_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12960256)))]; |
| 222 | tensor<fp16, [1, 768, 1, 1500]> var_481_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_481_dilations_0, groups = var_481_groups_0, pad = var_481_pad_0, pad_type = var_481_pad_type_0, strides = var_481_strides_0, weight = layers_1_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_5_cast_fp16)[name = tensor<string, []>("op_481_cast_fp16")]; |
| 223 | tensor<string, []> var_487_pad_type_0 = const()[name = tensor<string, []>("op_487_pad_type_0"), val = tensor<string, []>("valid")]; |
| 224 | tensor<int32, [2]> var_487_strides_0 = const()[name = tensor<string, []>("op_487_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 225 | tensor<int32, [4]> var_487_pad_0 = const()[name = tensor<string, []>("op_487_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 226 | tensor<int32, [2]> var_487_dilations_0 = const()[name = tensor<string, []>("op_487_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 227 | tensor<int32, []> var_487_groups_0 = const()[name = tensor<string, []>("op_487_groups_0"), val = tensor<int32, []>(1)]; |
| 228 | tensor<fp16, [768, 768, 1, 1]> layers_1_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12975872))), name = tensor<string, []>("layers_1_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [6964]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12961856))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 229 | tensor<fp16, [1, 768, 1, 1500]> var_487_cast_fp16 = conv(dilations = var_487_dilations_0, groups = var_487_groups_0, pad = var_487_pad_0, pad_type = var_487_pad_type_0, strides = var_487_strides_0, weight = layers_1_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_5_cast_fp16)[name = tensor<string, []>("op_487_cast_fp16")]; |
| 230 | tensor<fp16, [1, 768, 1, 1500]> value_3_cast_fp16 = add(x = var_481_cast_fp16, y = var_487_cast_fp16)[name = tensor<string, []>("value_3_cast_fp16")]; |
| 231 | tensor<int32, [4]> var_491 = const()[name = tensor<string, []>("op_491"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 232 | tensor<fp16, [1, 12, 64, 1500]> mh_q_3_cast_fp16 = reshape(shape = var_491, x = query_3_cast_fp16)[name = tensor<string, []>("mh_q_3_cast_fp16")]; |
| 233 | tensor<fp16, []> var_493_to_fp16 = const()[name = tensor<string, []>("op_493_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| 234 | tensor<fp16, [1, 12, 64, 1500]> var_494_cast_fp16 = mul(x = mh_q_3_cast_fp16, y = var_493_to_fp16)[name = tensor<string, []>("op_494_cast_fp16")]; |
| 235 | tensor<int32, [4]> var_497 = const()[name = tensor<string, []>("op_497"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 236 | tensor<fp16, [1, 12, 64, 1500]> var_498_cast_fp16 = reshape(shape = var_497, x = key_3_cast_fp16)[name = tensor<string, []>("op_498_cast_fp16")]; |
| 237 | tensor<bool, []> mh_w_3_transpose_x_0 = const()[name = tensor<string, []>("mh_w_3_transpose_x_0"), val = tensor<bool, []>(true)]; |
| 238 | tensor<bool, []> mh_w_3_transpose_y_0 = const()[name = tensor<string, []>("mh_w_3_transpose_y_0"), val = tensor<bool, []>(false)]; |
| 239 | tensor<fp16, [1, 12, 1500, 1500]> mh_w_3_cast_fp16 = matmul(transpose_x = mh_w_3_transpose_x_0, transpose_y = mh_w_3_transpose_y_0, x = var_494_cast_fp16, y = var_498_cast_fp16)[name = tensor<string, []>("mh_w_3_cast_fp16")]; |
| 240 | tensor<fp16, [1, 12, 1500, 1500]> var_501_cast_fp16 = softmax(axis = var_406, x = mh_w_3_cast_fp16)[name = tensor<string, []>("op_501_cast_fp16")]; |
| 241 | tensor<int32, [4]> var_502 = const()[name = tensor<string, []>("op_502"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 242 | tensor<fp16, [1, 12, 64, 1500]> var_503_cast_fp16 = reshape(shape = var_502, x = value_3_cast_fp16)[name = tensor<string, []>("op_503_cast_fp16")]; |
| 243 | tensor<bool, []> attn_3_transpose_x_0 = const()[name = tensor<string, []>("attn_3_transpose_x_0"), val = tensor<bool, []>(false)]; |
| 244 | tensor<bool, []> attn_3_transpose_y_0 = const()[name = tensor<string, []>("attn_3_transpose_y_0"), val = tensor<bool, []>(true)]; |
| 245 | tensor<fp16, [1, 12, 64, 1500]> attn_3_cast_fp16 = matmul(transpose_x = attn_3_transpose_x_0, transpose_y = attn_3_transpose_y_0, x = var_503_cast_fp16, y = var_501_cast_fp16)[name = tensor<string, []>("attn_3_cast_fp16")]; |
| 246 | tensor<int32, [4]> var_506 = const()[name = tensor<string, []>("op_506"), val = tensor<int32, [4]>([1, 768, 1, 1500])]; |
| 247 | tensor<fp16, [1, 768, 1, 1500]> input_9_cast_fp16 = reshape(shape = var_506, x = attn_3_cast_fp16)[name = tensor<string, []>("input_9_cast_fp16")]; |
| 248 | tensor<string, []> var_516_pad_type_0 = const()[name = tensor<string, []>("op_516_pad_type_0"), val = tensor<string, []>("valid")]; |
| 249 | tensor<int32, [2]> var_516_strides_0 = const()[name = tensor<string, []>("op_516_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 250 | tensor<int32, [4]> var_516_pad_0 = const()[name = tensor<string, []>("op_516_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 251 | tensor<int32, [2]> var_516_dilations_0 = const()[name = tensor<string, []>("op_516_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 252 | tensor<int32, []> var_516_groups_0 = const()[name = tensor<string, []>("op_516_groups_0"), val = tensor<int32, []>(1)]; |
| 253 | tensor<fp16, [768, 768, 1, 1]> layers_1_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13049664))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13344640))), name = tensor<string, []>("layers_1_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 254 | tensor<fp16, [768]> layers_1_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13344768)))]; |
| 255 | tensor<fp16, [1, 768, 1, 1500]> var_516_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_516_dilations_0, groups = var_516_groups_0, pad = var_516_pad_0, pad_type = var_516_pad_type_0, strides = var_516_strides_0, weight = layers_1_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_9_cast_fp16)[name = tensor<string, []>("op_516_cast_fp16")]; |
| 256 | tensor<string, []> var_522_pad_type_0 = const()[name = tensor<string, []>("op_522_pad_type_0"), val = tensor<string, []>("valid")]; |
| 257 | tensor<int32, [2]> var_522_strides_0 = const()[name = tensor<string, []>("op_522_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 258 | tensor<int32, [4]> var_522_pad_0 = const()[name = tensor<string, []>("op_522_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 259 | tensor<int32, [2]> var_522_dilations_0 = const()[name = tensor<string, []>("op_522_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 260 | tensor<int32, []> var_522_groups_0 = const()[name = tensor<string, []>("op_522_groups_0"), val = tensor<int32, []>(1)]; |
| 261 | tensor<fp16, [768, 768, 1, 1]> layers_1_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13357056))), name = tensor<string, []>("layers_1_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [5299]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13346368))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 262 | tensor<fp16, [1, 768, 1, 1500]> var_522_cast_fp16 = conv(dilations = var_522_dilations_0, groups = var_522_groups_0, pad = var_522_pad_0, pad_type = var_522_pad_type_0, strides = var_522_strides_0, weight = layers_1_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_9_cast_fp16)[name = tensor<string, []>("op_522_cast_fp16")]; |
| 263 | tensor<fp16, [1, 768, 1, 1500]> obj_7_cast_fp16 = add(x = var_516_cast_fp16, y = var_522_cast_fp16)[name = tensor<string, []>("obj_7_cast_fp16")]; |
| 264 | tensor<fp16, [1, 768, 1, 1500]> inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = obj_7_cast_fp16)[name = tensor<string, []>("inputs_7_cast_fp16")]; |
| 265 | tensor<int32, [1]> out_7_axes_0 = const()[name = tensor<string, []>("out_7_axes_0"), val = tensor<int32, [1]>([1])]; |
| 266 | tensor<fp16, []> var_533_to_fp16 = const()[name = tensor<string, []>("op_533_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 267 | tensor<fp16, [1, 768, 1, 1500]> out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_533_to_fp16, x = inputs_7_cast_fp16)[name = tensor<string, []>("out_7_cast_fp16")]; |
| 268 | tensor<fp16, [768]> input_11_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_11_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13430848)))]; |
| 269 | tensor<fp16, [768]> input_11_beta_0_to_fp16 = const()[name = tensor<string, []>("input_11_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13432448)))]; |
| 270 | tensor<fp16, []> input_11_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_11_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 271 | tensor<fp16, [1, 768, 1, 1500]> input_11_cast_fp16 = batch_norm(beta = input_11_beta_0_to_fp16, epsilon = input_11_epsilon_0_to_fp16, gamma = input_11_gamma_0_to_fp16, mean = var_57_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_7_cast_fp16)[name = tensor<string, []>("input_11_cast_fp16")]; |
| 272 | tensor<string, []> var_551_pad_type_0 = const()[name = tensor<string, []>("op_551_pad_type_0"), val = tensor<string, []>("valid")]; |
| 273 | tensor<int32, [2]> var_551_strides_0 = const()[name = tensor<string, []>("op_551_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 274 | tensor<int32, [4]> var_551_pad_0 = const()[name = tensor<string, []>("op_551_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 275 | tensor<int32, [2]> var_551_dilations_0 = const()[name = tensor<string, []>("op_551_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 276 | tensor<int32, []> var_551_groups_0 = const()[name = tensor<string, []>("op_551_groups_0"), val = tensor<int32, []>(1)]; |
| 277 | tensor<fp16, [3072, 768, 1, 1]> layers_1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1179648]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13434048))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14613760))), name = tensor<string, []>("layers_1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([3072, 768, 1, 1])]; |
| 278 | tensor<fp16, [3072]> layers_1_fc1_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_fc1_inlier_module_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14613888)))]; |
| 279 | tensor<fp16, [1, 3072, 1, 1500]> var_551_cast_fp16 = conv(bias = layers_1_fc1_inlier_module_bias_to_fp16, dilations = var_551_dilations_0, groups = var_551_groups_0, pad = var_551_pad_0, pad_type = var_551_pad_type_0, strides = var_551_strides_0, weight = layers_1_fc1_inlier_module_weight_to_fp16_palettized, x = input_11_cast_fp16)[name = tensor<string, []>("op_551_cast_fp16")]; |
| 280 | tensor<string, []> var_557_pad_type_0 = const()[name = tensor<string, []>("op_557_pad_type_0"), val = tensor<string, []>("valid")]; |
| 281 | tensor<int32, [2]> var_557_strides_0 = const()[name = tensor<string, []>("op_557_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 282 | tensor<int32, [4]> var_557_pad_0 = const()[name = tensor<string, []>("op_557_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 283 | tensor<int32, [2]> var_557_dilations_0 = const()[name = tensor<string, []>("op_557_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 284 | tensor<int32, []> var_557_groups_0 = const()[name = tensor<string, []>("op_557_groups_0"), val = tensor<int32, []>(1)]; |
| 285 | tensor<fp16, [3072, 768, 1, 1]> layers_1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14688448))), name = tensor<string, []>("layers_1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [34117]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14620096))), shape = tensor<uint32, [4]>([3072, 768, 1, 1])]; |
| 286 | tensor<fp16, [1, 3072, 1, 1500]> var_557_cast_fp16 = conv(dilations = var_557_dilations_0, groups = var_557_groups_0, pad = var_557_pad_0, pad_type = var_557_pad_type_0, strides = var_557_strides_0, weight = layers_1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_11_cast_fp16)[name = tensor<string, []>("op_557_cast_fp16")]; |
| 287 | tensor<fp16, [1, 3072, 1, 1500]> input_13_cast_fp16 = add(x = var_551_cast_fp16, y = var_557_cast_fp16)[name = tensor<string, []>("input_13_cast_fp16")]; |
| 288 | tensor<string, []> input_15_mode_0 = const()[name = tensor<string, []>("input_15_mode_0"), val = tensor<string, []>("EXACT")]; |
| 289 | tensor<fp16, [1, 3072, 1, 1500]> input_15_cast_fp16 = gelu(mode = input_15_mode_0, x = input_13_cast_fp16)[name = tensor<string, []>("input_15_cast_fp16")]; |
| 290 | tensor<string, []> var_568_pad_type_0 = const()[name = tensor<string, []>("op_568_pad_type_0"), val = tensor<string, []>("valid")]; |
| 291 | tensor<int32, [2]> var_568_strides_0 = const()[name = tensor<string, []>("op_568_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 292 | tensor<int32, [4]> var_568_pad_0 = const()[name = tensor<string, []>("op_568_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 293 | tensor<int32, [2]> var_568_dilations_0 = const()[name = tensor<string, []>("op_568_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 294 | tensor<int32, []> var_568_groups_0 = const()[name = tensor<string, []>("op_568_groups_0"), val = tensor<int32, []>(1)]; |
| 295 | tensor<fp16, [768, 3072, 1, 1]> layers_1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1179648]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14983424))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16163136))), name = tensor<string, []>("layers_1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 3072, 1, 1])]; |
| 296 | tensor<fp16, [768]> layers_1_fc2_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_fc2_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16163264)))]; |
| 297 | tensor<fp16, [1, 768, 1, 1500]> var_568_cast_fp16 = conv(bias = layers_1_fc2_inlier_module_bias_to_fp16, dilations = var_568_dilations_0, groups = var_568_groups_0, pad = var_568_pad_0, pad_type = var_568_pad_type_0, strides = var_568_strides_0, weight = layers_1_fc2_inlier_module_weight_to_fp16_palettized, x = input_15_cast_fp16)[name = tensor<string, []>("op_568_cast_fp16")]; |
| 298 | tensor<string, []> var_574_pad_type_0 = const()[name = tensor<string, []>("op_574_pad_type_0"), val = tensor<string, []>("valid")]; |
| 299 | tensor<int32, [2]> var_574_strides_0 = const()[name = tensor<string, []>("op_574_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 300 | tensor<int32, [4]> var_574_pad_0 = const()[name = tensor<string, []>("op_574_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 301 | tensor<int32, [2]> var_574_dilations_0 = const()[name = tensor<string, []>("op_574_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 302 | tensor<int32, []> var_574_groups_0 = const()[name = tensor<string, []>("op_574_groups_0"), val = tensor<int32, []>(1)]; |
| 303 | tensor<fp16, [768, 3072, 1, 1]> layers_1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16225152))), name = tensor<string, []>("layers_1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [30100]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16164864))), shape = tensor<uint32, [4]>([768, 3072, 1, 1])]; |
| 304 | tensor<fp16, [1, 768, 1, 1500]> var_574_cast_fp16 = conv(dilations = var_574_dilations_0, groups = var_574_groups_0, pad = var_574_pad_0, pad_type = var_574_pad_type_0, strides = var_574_strides_0, weight = layers_1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_15_cast_fp16)[name = tensor<string, []>("op_574_cast_fp16")]; |
| 305 | tensor<fp16, [1, 768, 1, 1500]> hidden_states_7_cast_fp16 = add(x = var_568_cast_fp16, y = var_574_cast_fp16)[name = tensor<string, []>("hidden_states_7_cast_fp16")]; |
| 306 | tensor<fp16, [1, 768, 1, 1500]> inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = hidden_states_7_cast_fp16)[name = tensor<string, []>("inputs_9_cast_fp16")]; |
| 307 | tensor<int32, []> var_580 = const()[name = tensor<string, []>("op_580"), val = tensor<int32, []>(3)]; |
| 308 | tensor<int32, [1]> out_9_axes_0 = const()[name = tensor<string, []>("out_9_axes_0"), val = tensor<int32, [1]>([1])]; |
| 309 | tensor<fp16, []> var_602_to_fp16 = const()[name = tensor<string, []>("op_602_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 310 | tensor<fp16, [1, 768, 1, 1500]> out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_602_to_fp16, x = inputs_9_cast_fp16)[name = tensor<string, []>("out_9_cast_fp16")]; |
| 311 | tensor<fp16, [768]> obj_9_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_9_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16520128)))]; |
| 312 | tensor<fp16, [768]> obj_9_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_9_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16521728)))]; |
| 313 | tensor<fp16, []> obj_9_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_9_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 314 | tensor<fp16, [1, 768, 1, 1500]> obj_9_cast_fp16 = batch_norm(beta = obj_9_beta_0_to_fp16, epsilon = obj_9_epsilon_0_to_fp16, gamma = obj_9_gamma_0_to_fp16, mean = var_57_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_9_cast_fp16)[name = tensor<string, []>("obj_9_cast_fp16")]; |
| 315 | tensor<string, []> var_624_pad_type_0 = const()[name = tensor<string, []>("op_624_pad_type_0"), val = tensor<string, []>("valid")]; |
| 316 | tensor<int32, [2]> var_624_strides_0 = const()[name = tensor<string, []>("op_624_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 317 | tensor<int32, [4]> var_624_pad_0 = const()[name = tensor<string, []>("op_624_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 318 | tensor<int32, [2]> var_624_dilations_0 = const()[name = tensor<string, []>("op_624_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 319 | tensor<int32, []> var_624_groups_0 = const()[name = tensor<string, []>("op_624_groups_0"), val = tensor<int32, []>(1)]; |
| 320 | tensor<fp16, [768, 768, 1, 1]> layers_2_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16523328))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16818304))), name = tensor<string, []>("layers_2_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 321 | tensor<fp16, [768]> layers_2_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16818432)))]; |
| 322 | tensor<fp16, [1, 768, 1, 1500]> var_624_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_624_dilations_0, groups = var_624_groups_0, pad = var_624_pad_0, pad_type = var_624_pad_type_0, strides = var_624_strides_0, weight = layers_2_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_9_cast_fp16)[name = tensor<string, []>("op_624_cast_fp16")]; |
| 323 | tensor<string, []> var_630_pad_type_0 = const()[name = tensor<string, []>("op_630_pad_type_0"), val = tensor<string, []>("valid")]; |
| 324 | tensor<int32, [2]> var_630_strides_0 = const()[name = tensor<string, []>("op_630_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 325 | tensor<int32, [4]> var_630_pad_0 = const()[name = tensor<string, []>("op_630_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 326 | tensor<int32, [2]> var_630_dilations_0 = const()[name = tensor<string, []>("op_630_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 327 | tensor<int32, []> var_630_groups_0 = const()[name = tensor<string, []>("op_630_groups_0"), val = tensor<int32, []>(1)]; |
| 328 | tensor<fp16, [768, 768, 1, 1]> layers_2_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16834112))), name = tensor<string, []>("layers_2_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [6978]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16820032))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 329 | tensor<fp16, [1, 768, 1, 1500]> var_630_cast_fp16 = conv(dilations = var_630_dilations_0, groups = var_630_groups_0, pad = var_630_pad_0, pad_type = var_630_pad_type_0, strides = var_630_strides_0, weight = layers_2_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_9_cast_fp16)[name = tensor<string, []>("op_630_cast_fp16")]; |
| 330 | tensor<fp16, [1, 768, 1, 1500]> query_5_cast_fp16 = add(x = var_624_cast_fp16, y = var_630_cast_fp16)[name = tensor<string, []>("query_5_cast_fp16")]; |
| 331 | tensor<string, []> var_639_pad_type_0 = const()[name = tensor<string, []>("op_639_pad_type_0"), val = tensor<string, []>("valid")]; |
| 332 | tensor<int32, [2]> var_639_strides_0 = const()[name = tensor<string, []>("op_639_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 333 | tensor<int32, [4]> var_639_pad_0 = const()[name = tensor<string, []>("op_639_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 334 | tensor<int32, [2]> var_639_dilations_0 = const()[name = tensor<string, []>("op_639_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 335 | tensor<int32, []> var_639_groups_0 = const()[name = tensor<string, []>("op_639_groups_0"), val = tensor<int32, []>(1)]; |
| 336 | tensor<fp16, [768, 768, 1, 1]> layers_2_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16907904))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17202880))), name = tensor<string, []>("layers_2_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 337 | tensor<fp16, [1, 768, 1, 1500]> var_639_cast_fp16 = conv(dilations = var_639_dilations_0, groups = var_639_groups_0, pad = var_639_pad_0, pad_type = var_639_pad_type_0, strides = var_639_strides_0, weight = layers_2_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_9_cast_fp16)[name = tensor<string, []>("op_639_cast_fp16")]; |
| 338 | tensor<string, []> var_645_pad_type_0 = const()[name = tensor<string, []>("op_645_pad_type_0"), val = tensor<string, []>("valid")]; |
| 339 | tensor<int32, [2]> var_645_strides_0 = const()[name = tensor<string, []>("op_645_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 340 | tensor<int32, [4]> var_645_pad_0 = const()[name = tensor<string, []>("op_645_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 341 | tensor<int32, [2]> var_645_dilations_0 = const()[name = tensor<string, []>("op_645_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 342 | tensor<int32, []> var_645_groups_0 = const()[name = tensor<string, []>("op_645_groups_0"), val = tensor<int32, []>(1)]; |
| 343 | tensor<fp16, [768, 768, 1, 1]> layers_2_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17216384))), name = tensor<string, []>("layers_2_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [6646]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17203008))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 344 | tensor<fp16, [1, 768, 1, 1500]> var_645_cast_fp16 = conv(dilations = var_645_dilations_0, groups = var_645_groups_0, pad = var_645_pad_0, pad_type = var_645_pad_type_0, strides = var_645_strides_0, weight = layers_2_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_9_cast_fp16)[name = tensor<string, []>("op_645_cast_fp16")]; |
| 345 | tensor<fp16, [1, 768, 1, 1500]> key_5_cast_fp16 = add(x = var_639_cast_fp16, y = var_645_cast_fp16)[name = tensor<string, []>("key_5_cast_fp16")]; |
| 346 | tensor<string, []> var_655_pad_type_0 = const()[name = tensor<string, []>("op_655_pad_type_0"), val = tensor<string, []>("valid")]; |
| 347 | tensor<int32, [2]> var_655_strides_0 = const()[name = tensor<string, []>("op_655_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 348 | tensor<int32, [4]> var_655_pad_0 = const()[name = tensor<string, []>("op_655_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 349 | tensor<int32, [2]> var_655_dilations_0 = const()[name = tensor<string, []>("op_655_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 350 | tensor<int32, []> var_655_groups_0 = const()[name = tensor<string, []>("op_655_groups_0"), val = tensor<int32, []>(1)]; |
| 351 | tensor<fp16, [768, 768, 1, 1]> layers_2_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17290176))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17585152))), name = tensor<string, []>("layers_2_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 352 | tensor<fp16, [768]> layers_2_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17585280)))]; |
| 353 | tensor<fp16, [1, 768, 1, 1500]> var_655_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_655_dilations_0, groups = var_655_groups_0, pad = var_655_pad_0, pad_type = var_655_pad_type_0, strides = var_655_strides_0, weight = layers_2_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_9_cast_fp16)[name = tensor<string, []>("op_655_cast_fp16")]; |
| 354 | tensor<string, []> var_661_pad_type_0 = const()[name = tensor<string, []>("op_661_pad_type_0"), val = tensor<string, []>("valid")]; |
| 355 | tensor<int32, [2]> var_661_strides_0 = const()[name = tensor<string, []>("op_661_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 356 | tensor<int32, [4]> var_661_pad_0 = const()[name = tensor<string, []>("op_661_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 357 | tensor<int32, [2]> var_661_dilations_0 = const()[name = tensor<string, []>("op_661_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 358 | tensor<int32, []> var_661_groups_0 = const()[name = tensor<string, []>("op_661_groups_0"), val = tensor<int32, []>(1)]; |
| 359 | tensor<fp16, [768, 768, 1, 1]> layers_2_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17596992))), name = tensor<string, []>("layers_2_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [5006]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17586880))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 360 | tensor<fp16, [1, 768, 1, 1500]> var_661_cast_fp16 = conv(dilations = var_661_dilations_0, groups = var_661_groups_0, pad = var_661_pad_0, pad_type = var_661_pad_type_0, strides = var_661_strides_0, weight = layers_2_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_9_cast_fp16)[name = tensor<string, []>("op_661_cast_fp16")]; |
| 361 | tensor<fp16, [1, 768, 1, 1500]> value_5_cast_fp16 = add(x = var_655_cast_fp16, y = var_661_cast_fp16)[name = tensor<string, []>("value_5_cast_fp16")]; |
| 362 | tensor<int32, [4]> var_665 = const()[name = tensor<string, []>("op_665"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 363 | tensor<fp16, [1, 12, 64, 1500]> mh_q_5_cast_fp16 = reshape(shape = var_665, x = query_5_cast_fp16)[name = tensor<string, []>("mh_q_5_cast_fp16")]; |
| 364 | tensor<fp16, []> var_667_to_fp16 = const()[name = tensor<string, []>("op_667_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| 365 | tensor<fp16, [1, 12, 64, 1500]> var_668_cast_fp16 = mul(x = mh_q_5_cast_fp16, y = var_667_to_fp16)[name = tensor<string, []>("op_668_cast_fp16")]; |
| 366 | tensor<int32, [4]> var_671 = const()[name = tensor<string, []>("op_671"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 367 | tensor<fp16, [1, 12, 64, 1500]> var_672_cast_fp16 = reshape(shape = var_671, x = key_5_cast_fp16)[name = tensor<string, []>("op_672_cast_fp16")]; |
| 368 | tensor<bool, []> mh_w_5_transpose_x_0 = const()[name = tensor<string, []>("mh_w_5_transpose_x_0"), val = tensor<bool, []>(true)]; |
| 369 | tensor<bool, []> mh_w_5_transpose_y_0 = const()[name = tensor<string, []>("mh_w_5_transpose_y_0"), val = tensor<bool, []>(false)]; |
| 370 | tensor<fp16, [1, 12, 1500, 1500]> mh_w_5_cast_fp16 = matmul(transpose_x = mh_w_5_transpose_x_0, transpose_y = mh_w_5_transpose_y_0, x = var_668_cast_fp16, y = var_672_cast_fp16)[name = tensor<string, []>("mh_w_5_cast_fp16")]; |
| 371 | tensor<fp16, [1, 12, 1500, 1500]> var_675_cast_fp16 = softmax(axis = var_580, x = mh_w_5_cast_fp16)[name = tensor<string, []>("op_675_cast_fp16")]; |
| 372 | tensor<int32, [4]> var_676 = const()[name = tensor<string, []>("op_676"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 373 | tensor<fp16, [1, 12, 64, 1500]> var_677_cast_fp16 = reshape(shape = var_676, x = value_5_cast_fp16)[name = tensor<string, []>("op_677_cast_fp16")]; |
| 374 | tensor<bool, []> attn_5_transpose_x_0 = const()[name = tensor<string, []>("attn_5_transpose_x_0"), val = tensor<bool, []>(false)]; |
| 375 | tensor<bool, []> attn_5_transpose_y_0 = const()[name = tensor<string, []>("attn_5_transpose_y_0"), val = tensor<bool, []>(true)]; |
| 376 | tensor<fp16, [1, 12, 64, 1500]> attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_677_cast_fp16, y = var_675_cast_fp16)[name = tensor<string, []>("attn_5_cast_fp16")]; |
| 377 | tensor<int32, [4]> var_680 = const()[name = tensor<string, []>("op_680"), val = tensor<int32, [4]>([1, 768, 1, 1500])]; |
| 378 | tensor<fp16, [1, 768, 1, 1500]> input_17_cast_fp16 = reshape(shape = var_680, x = attn_5_cast_fp16)[name = tensor<string, []>("input_17_cast_fp16")]; |
| 379 | tensor<string, []> var_690_pad_type_0 = const()[name = tensor<string, []>("op_690_pad_type_0"), val = tensor<string, []>("valid")]; |
| 380 | tensor<int32, [2]> var_690_strides_0 = const()[name = tensor<string, []>("op_690_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 381 | tensor<int32, [4]> var_690_pad_0 = const()[name = tensor<string, []>("op_690_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 382 | tensor<int32, [2]> var_690_dilations_0 = const()[name = tensor<string, []>("op_690_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 383 | tensor<int32, []> var_690_groups_0 = const()[name = tensor<string, []>("op_690_groups_0"), val = tensor<int32, []>(1)]; |
| 384 | tensor<fp16, [768, 768, 1, 1]> layers_2_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17670784))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17965760))), name = tensor<string, []>("layers_2_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 385 | tensor<fp16, [768]> layers_2_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17965888)))]; |
| 386 | tensor<fp16, [1, 768, 1, 1500]> var_690_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_690_dilations_0, groups = var_690_groups_0, pad = var_690_pad_0, pad_type = var_690_pad_type_0, strides = var_690_strides_0, weight = layers_2_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_17_cast_fp16)[name = tensor<string, []>("op_690_cast_fp16")]; |
| 387 | tensor<string, []> var_696_pad_type_0 = const()[name = tensor<string, []>("op_696_pad_type_0"), val = tensor<string, []>("valid")]; |
| 388 | tensor<int32, [2]> var_696_strides_0 = const()[name = tensor<string, []>("op_696_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 389 | tensor<int32, [4]> var_696_pad_0 = const()[name = tensor<string, []>("op_696_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 390 | tensor<int32, [2]> var_696_dilations_0 = const()[name = tensor<string, []>("op_696_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 391 | tensor<int32, []> var_696_groups_0 = const()[name = tensor<string, []>("op_696_groups_0"), val = tensor<int32, []>(1)]; |
| 392 | tensor<fp16, [768, 768, 1, 1]> layers_2_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17976064))), name = tensor<string, []>("layers_2_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [4255]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17967488))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 393 | tensor<fp16, [1, 768, 1, 1500]> var_696_cast_fp16 = conv(dilations = var_696_dilations_0, groups = var_696_groups_0, pad = var_696_pad_0, pad_type = var_696_pad_type_0, strides = var_696_strides_0, weight = layers_2_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_17_cast_fp16)[name = tensor<string, []>("op_696_cast_fp16")]; |
| 394 | tensor<fp16, [1, 768, 1, 1500]> obj_11_cast_fp16 = add(x = var_690_cast_fp16, y = var_696_cast_fp16)[name = tensor<string, []>("obj_11_cast_fp16")]; |
| 395 | tensor<fp16, [1, 768, 1, 1500]> inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_11_cast_fp16)[name = tensor<string, []>("inputs_11_cast_fp16")]; |
| 396 | tensor<int32, [1]> out_11_axes_0 = const()[name = tensor<string, []>("out_11_axes_0"), val = tensor<int32, [1]>([1])]; |
| 397 | tensor<fp16, []> var_707_to_fp16 = const()[name = tensor<string, []>("op_707_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 398 | tensor<fp16, [1, 768, 1, 1500]> out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_707_to_fp16, x = inputs_11_cast_fp16)[name = tensor<string, []>("out_11_cast_fp16")]; |
| 399 | tensor<fp16, [768]> input_19_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_19_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18049856)))]; |
| 400 | tensor<fp16, [768]> input_19_beta_0_to_fp16 = const()[name = tensor<string, []>("input_19_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18051456)))]; |
| 401 | tensor<fp16, []> input_19_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_19_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 402 | tensor<fp16, [1, 768, 1, 1500]> input_19_cast_fp16 = batch_norm(beta = input_19_beta_0_to_fp16, epsilon = input_19_epsilon_0_to_fp16, gamma = input_19_gamma_0_to_fp16, mean = var_57_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_11_cast_fp16)[name = tensor<string, []>("input_19_cast_fp16")]; |
| 403 | tensor<string, []> var_725_pad_type_0 = const()[name = tensor<string, []>("op_725_pad_type_0"), val = tensor<string, []>("valid")]; |
| 404 | tensor<int32, [2]> var_725_strides_0 = const()[name = tensor<string, []>("op_725_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 405 | tensor<int32, [4]> var_725_pad_0 = const()[name = tensor<string, []>("op_725_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 406 | tensor<int32, [2]> var_725_dilations_0 = const()[name = tensor<string, []>("op_725_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 407 | tensor<int32, []> var_725_groups_0 = const()[name = tensor<string, []>("op_725_groups_0"), val = tensor<int32, []>(1)]; |
| 408 | tensor<fp16, [3072, 768, 1, 1]> layers_2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1179648]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18053056))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19232768))), name = tensor<string, []>("layers_2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([3072, 768, 1, 1])]; |
| 409 | tensor<fp16, [3072]> layers_2_fc1_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_fc1_inlier_module_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19232896)))]; |
| 410 | tensor<fp16, [1, 3072, 1, 1500]> var_725_cast_fp16 = conv(bias = layers_2_fc1_inlier_module_bias_to_fp16, dilations = var_725_dilations_0, groups = var_725_groups_0, pad = var_725_pad_0, pad_type = var_725_pad_type_0, strides = var_725_strides_0, weight = layers_2_fc1_inlier_module_weight_to_fp16_palettized, x = input_19_cast_fp16)[name = tensor<string, []>("op_725_cast_fp16")]; |
| 411 | tensor<string, []> var_731_pad_type_0 = const()[name = tensor<string, []>("op_731_pad_type_0"), val = tensor<string, []>("valid")]; |
| 412 | tensor<int32, [2]> var_731_strides_0 = const()[name = tensor<string, []>("op_731_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 413 | tensor<int32, [4]> var_731_pad_0 = const()[name = tensor<string, []>("op_731_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 414 | tensor<int32, [2]> var_731_dilations_0 = const()[name = tensor<string, []>("op_731_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 415 | tensor<int32, []> var_731_groups_0 = const()[name = tensor<string, []>("op_731_groups_0"), val = tensor<int32, []>(1)]; |
| 416 | tensor<fp16, [3072, 768, 1, 1]> layers_2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19298752))), name = tensor<string, []>("layers_2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [29780]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19239104))), shape = tensor<uint32, [4]>([3072, 768, 1, 1])]; |
| 417 | tensor<fp16, [1, 3072, 1, 1500]> var_731_cast_fp16 = conv(dilations = var_731_dilations_0, groups = var_731_groups_0, pad = var_731_pad_0, pad_type = var_731_pad_type_0, strides = var_731_strides_0, weight = layers_2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_19_cast_fp16)[name = tensor<string, []>("op_731_cast_fp16")]; |
| 418 | tensor<fp16, [1, 3072, 1, 1500]> input_21_cast_fp16 = add(x = var_725_cast_fp16, y = var_731_cast_fp16)[name = tensor<string, []>("input_21_cast_fp16")]; |
| 419 | tensor<string, []> input_23_mode_0 = const()[name = tensor<string, []>("input_23_mode_0"), val = tensor<string, []>("EXACT")]; |
| 420 | tensor<fp16, [1, 3072, 1, 1500]> input_23_cast_fp16 = gelu(mode = input_23_mode_0, x = input_21_cast_fp16)[name = tensor<string, []>("input_23_cast_fp16")]; |
| 421 | tensor<string, []> var_742_pad_type_0 = const()[name = tensor<string, []>("op_742_pad_type_0"), val = tensor<string, []>("valid")]; |
| 422 | tensor<int32, [2]> var_742_strides_0 = const()[name = tensor<string, []>("op_742_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 423 | tensor<int32, [4]> var_742_pad_0 = const()[name = tensor<string, []>("op_742_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 424 | tensor<int32, [2]> var_742_dilations_0 = const()[name = tensor<string, []>("op_742_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 425 | tensor<int32, []> var_742_groups_0 = const()[name = tensor<string, []>("op_742_groups_0"), val = tensor<int32, []>(1)]; |
| 426 | tensor<fp16, [768, 3072, 1, 1]> layers_2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1179648]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19593728))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20773440))), name = tensor<string, []>("layers_2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 3072, 1, 1])]; |
| 427 | tensor<fp16, [768]> layers_2_fc2_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_2_fc2_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20773568)))]; |
| 428 | tensor<fp16, [1, 768, 1, 1500]> var_742_cast_fp16 = conv(bias = layers_2_fc2_inlier_module_bias_to_fp16, dilations = var_742_dilations_0, groups = var_742_groups_0, pad = var_742_pad_0, pad_type = var_742_pad_type_0, strides = var_742_strides_0, weight = layers_2_fc2_inlier_module_weight_to_fp16_palettized, x = input_23_cast_fp16)[name = tensor<string, []>("op_742_cast_fp16")]; |
| 429 | tensor<string, []> var_748_pad_type_0 = const()[name = tensor<string, []>("op_748_pad_type_0"), val = tensor<string, []>("valid")]; |
| 430 | tensor<int32, [2]> var_748_strides_0 = const()[name = tensor<string, []>("op_748_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 431 | tensor<int32, [4]> var_748_pad_0 = const()[name = tensor<string, []>("op_748_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 432 | tensor<int32, [2]> var_748_dilations_0 = const()[name = tensor<string, []>("op_748_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 433 | tensor<int32, []> var_748_groups_0 = const()[name = tensor<string, []>("op_748_groups_0"), val = tensor<int32, []>(1)]; |
| 434 | tensor<fp16, [768, 3072, 1, 1]> layers_2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20834944))), name = tensor<string, []>("layers_2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [29841]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20775168))), shape = tensor<uint32, [4]>([768, 3072, 1, 1])]; |
| 435 | tensor<fp16, [1, 768, 1, 1500]> var_748_cast_fp16 = conv(dilations = var_748_dilations_0, groups = var_748_groups_0, pad = var_748_pad_0, pad_type = var_748_pad_type_0, strides = var_748_strides_0, weight = layers_2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_23_cast_fp16)[name = tensor<string, []>("op_748_cast_fp16")]; |
| 436 | tensor<fp16, [1, 768, 1, 1500]> hidden_states_9_cast_fp16 = add(x = var_742_cast_fp16, y = var_748_cast_fp16)[name = tensor<string, []>("hidden_states_9_cast_fp16")]; |
| 437 | tensor<fp16, [1, 768, 1, 1500]> inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_9_cast_fp16)[name = tensor<string, []>("inputs_13_cast_fp16")]; |
| 438 | tensor<int32, []> var_754 = const()[name = tensor<string, []>("op_754"), val = tensor<int32, []>(3)]; |
| 439 | tensor<int32, [1]> out_13_axes_0 = const()[name = tensor<string, []>("out_13_axes_0"), val = tensor<int32, [1]>([1])]; |
| 440 | tensor<fp16, []> var_776_to_fp16 = const()[name = tensor<string, []>("op_776_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 441 | tensor<fp16, [1, 768, 1, 1500]> out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_776_to_fp16, x = inputs_13_cast_fp16)[name = tensor<string, []>("out_13_cast_fp16")]; |
| 442 | tensor<fp16, [768]> obj_13_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_13_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21129920)))]; |
| 443 | tensor<fp16, [768]> obj_13_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_13_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21131520)))]; |
| 444 | tensor<fp16, []> obj_13_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_13_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 445 | tensor<fp16, [1, 768, 1, 1500]> obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = var_57_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_13_cast_fp16)[name = tensor<string, []>("obj_13_cast_fp16")]; |
| 446 | tensor<string, []> var_798_pad_type_0 = const()[name = tensor<string, []>("op_798_pad_type_0"), val = tensor<string, []>("valid")]; |
| 447 | tensor<int32, [2]> var_798_strides_0 = const()[name = tensor<string, []>("op_798_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 448 | tensor<int32, [4]> var_798_pad_0 = const()[name = tensor<string, []>("op_798_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 449 | tensor<int32, [2]> var_798_dilations_0 = const()[name = tensor<string, []>("op_798_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 450 | tensor<int32, []> var_798_groups_0 = const()[name = tensor<string, []>("op_798_groups_0"), val = tensor<int32, []>(1)]; |
| 451 | tensor<fp16, [768, 768, 1, 1]> layers_3_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21133120))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21428096))), name = tensor<string, []>("layers_3_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 452 | tensor<fp16, [768]> layers_3_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21428224)))]; |
| 453 | tensor<fp16, [1, 768, 1, 1500]> var_798_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_798_dilations_0, groups = var_798_groups_0, pad = var_798_pad_0, pad_type = var_798_pad_type_0, strides = var_798_strides_0, weight = layers_3_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_13_cast_fp16)[name = tensor<string, []>("op_798_cast_fp16")]; |
| 454 | tensor<string, []> var_804_pad_type_0 = const()[name = tensor<string, []>("op_804_pad_type_0"), val = tensor<string, []>("valid")]; |
| 455 | tensor<int32, [2]> var_804_strides_0 = const()[name = tensor<string, []>("op_804_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 456 | tensor<int32, [4]> var_804_pad_0 = const()[name = tensor<string, []>("op_804_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 457 | tensor<int32, [2]> var_804_dilations_0 = const()[name = tensor<string, []>("op_804_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 458 | tensor<int32, []> var_804_groups_0 = const()[name = tensor<string, []>("op_804_groups_0"), val = tensor<int32, []>(1)]; |
| 459 | tensor<fp16, [768, 768, 1, 1]> layers_3_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21440384))), name = tensor<string, []>("layers_3_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [5218]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21429824))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 460 | tensor<fp16, [1, 768, 1, 1500]> var_804_cast_fp16 = conv(dilations = var_804_dilations_0, groups = var_804_groups_0, pad = var_804_pad_0, pad_type = var_804_pad_type_0, strides = var_804_strides_0, weight = layers_3_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_13_cast_fp16)[name = tensor<string, []>("op_804_cast_fp16")]; |
| 461 | tensor<fp16, [1, 768, 1, 1500]> query_7_cast_fp16 = add(x = var_798_cast_fp16, y = var_804_cast_fp16)[name = tensor<string, []>("query_7_cast_fp16")]; |
| 462 | tensor<string, []> var_813_pad_type_0 = const()[name = tensor<string, []>("op_813_pad_type_0"), val = tensor<string, []>("valid")]; |
| 463 | tensor<int32, [2]> var_813_strides_0 = const()[name = tensor<string, []>("op_813_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 464 | tensor<int32, [4]> var_813_pad_0 = const()[name = tensor<string, []>("op_813_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 465 | tensor<int32, [2]> var_813_dilations_0 = const()[name = tensor<string, []>("op_813_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 466 | tensor<int32, []> var_813_groups_0 = const()[name = tensor<string, []>("op_813_groups_0"), val = tensor<int32, []>(1)]; |
| 467 | tensor<fp16, [768, 768, 1, 1]> layers_3_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21514176))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21809152))), name = tensor<string, []>("layers_3_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 468 | tensor<fp16, [1, 768, 1, 1500]> var_813_cast_fp16 = conv(dilations = var_813_dilations_0, groups = var_813_groups_0, pad = var_813_pad_0, pad_type = var_813_pad_type_0, strides = var_813_strides_0, weight = layers_3_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_13_cast_fp16)[name = tensor<string, []>("op_813_cast_fp16")]; |
| 469 | tensor<string, []> var_819_pad_type_0 = const()[name = tensor<string, []>("op_819_pad_type_0"), val = tensor<string, []>("valid")]; |
| 470 | tensor<int32, [2]> var_819_strides_0 = const()[name = tensor<string, []>("op_819_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 471 | tensor<int32, [4]> var_819_pad_0 = const()[name = tensor<string, []>("op_819_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 472 | tensor<int32, [2]> var_819_dilations_0 = const()[name = tensor<string, []>("op_819_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 473 | tensor<int32, []> var_819_groups_0 = const()[name = tensor<string, []>("op_819_groups_0"), val = tensor<int32, []>(1)]; |
| 474 | tensor<fp16, [768, 768, 1, 1]> layers_3_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21819776))), name = tensor<string, []>("layers_3_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [5203]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21809280))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 475 | tensor<fp16, [1, 768, 1, 1500]> var_819_cast_fp16 = conv(dilations = var_819_dilations_0, groups = var_819_groups_0, pad = var_819_pad_0, pad_type = var_819_pad_type_0, strides = var_819_strides_0, weight = layers_3_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_13_cast_fp16)[name = tensor<string, []>("op_819_cast_fp16")]; |
| 476 | tensor<fp16, [1, 768, 1, 1500]> key_7_cast_fp16 = add(x = var_813_cast_fp16, y = var_819_cast_fp16)[name = tensor<string, []>("key_7_cast_fp16")]; |
| 477 | tensor<string, []> var_829_pad_type_0 = const()[name = tensor<string, []>("op_829_pad_type_0"), val = tensor<string, []>("valid")]; |
| 478 | tensor<int32, [2]> var_829_strides_0 = const()[name = tensor<string, []>("op_829_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 479 | tensor<int32, [4]> var_829_pad_0 = const()[name = tensor<string, []>("op_829_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 480 | tensor<int32, [2]> var_829_dilations_0 = const()[name = tensor<string, []>("op_829_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 481 | tensor<int32, []> var_829_groups_0 = const()[name = tensor<string, []>("op_829_groups_0"), val = tensor<int32, []>(1)]; |
| 482 | tensor<fp16, [768, 768, 1, 1]> layers_3_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21893568))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22188544))), name = tensor<string, []>("layers_3_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 483 | tensor<fp16, [768]> layers_3_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22188672)))]; |
| 484 | tensor<fp16, [1, 768, 1, 1500]> var_829_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_829_dilations_0, groups = var_829_groups_0, pad = var_829_pad_0, pad_type = var_829_pad_type_0, strides = var_829_strides_0, weight = layers_3_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_13_cast_fp16)[name = tensor<string, []>("op_829_cast_fp16")]; |
| 485 | tensor<string, []> var_835_pad_type_0 = const()[name = tensor<string, []>("op_835_pad_type_0"), val = tensor<string, []>("valid")]; |
| 486 | tensor<int32, [2]> var_835_strides_0 = const()[name = tensor<string, []>("op_835_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 487 | tensor<int32, [4]> var_835_pad_0 = const()[name = tensor<string, []>("op_835_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 488 | tensor<int32, [2]> var_835_dilations_0 = const()[name = tensor<string, []>("op_835_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 489 | tensor<int32, []> var_835_groups_0 = const()[name = tensor<string, []>("op_835_groups_0"), val = tensor<int32, []>(1)]; |
| 490 | tensor<fp16, [768, 768, 1, 1]> layers_3_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22198784))), name = tensor<string, []>("layers_3_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [4222]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22190272))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 491 | tensor<fp16, [1, 768, 1, 1500]> var_835_cast_fp16 = conv(dilations = var_835_dilations_0, groups = var_835_groups_0, pad = var_835_pad_0, pad_type = var_835_pad_type_0, strides = var_835_strides_0, weight = layers_3_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_13_cast_fp16)[name = tensor<string, []>("op_835_cast_fp16")]; |
| 492 | tensor<fp16, [1, 768, 1, 1500]> value_7_cast_fp16 = add(x = var_829_cast_fp16, y = var_835_cast_fp16)[name = tensor<string, []>("value_7_cast_fp16")]; |
| 493 | tensor<int32, [4]> var_839 = const()[name = tensor<string, []>("op_839"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 494 | tensor<fp16, [1, 12, 64, 1500]> mh_q_7_cast_fp16 = reshape(shape = var_839, x = query_7_cast_fp16)[name = tensor<string, []>("mh_q_7_cast_fp16")]; |
| 495 | tensor<fp16, []> var_841_to_fp16 = const()[name = tensor<string, []>("op_841_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| 496 | tensor<fp16, [1, 12, 64, 1500]> var_842_cast_fp16 = mul(x = mh_q_7_cast_fp16, y = var_841_to_fp16)[name = tensor<string, []>("op_842_cast_fp16")]; |
| 497 | tensor<int32, [4]> var_845 = const()[name = tensor<string, []>("op_845"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 498 | tensor<fp16, [1, 12, 64, 1500]> var_846_cast_fp16 = reshape(shape = var_845, x = key_7_cast_fp16)[name = tensor<string, []>("op_846_cast_fp16")]; |
| 499 | tensor<bool, []> mh_w_7_transpose_x_0 = const()[name = tensor<string, []>("mh_w_7_transpose_x_0"), val = tensor<bool, []>(true)]; |
| 500 | tensor<bool, []> mh_w_7_transpose_y_0 = const()[name = tensor<string, []>("mh_w_7_transpose_y_0"), val = tensor<bool, []>(false)]; |
| 501 | tensor<fp16, [1, 12, 1500, 1500]> mh_w_7_cast_fp16 = matmul(transpose_x = mh_w_7_transpose_x_0, transpose_y = mh_w_7_transpose_y_0, x = var_842_cast_fp16, y = var_846_cast_fp16)[name = tensor<string, []>("mh_w_7_cast_fp16")]; |
| 502 | tensor<fp16, [1, 12, 1500, 1500]> var_849_cast_fp16 = softmax(axis = var_754, x = mh_w_7_cast_fp16)[name = tensor<string, []>("op_849_cast_fp16")]; |
| 503 | tensor<int32, [4]> var_850 = const()[name = tensor<string, []>("op_850"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 504 | tensor<fp16, [1, 12, 64, 1500]> var_851_cast_fp16 = reshape(shape = var_850, x = value_7_cast_fp16)[name = tensor<string, []>("op_851_cast_fp16")]; |
| 505 | tensor<bool, []> attn_7_transpose_x_0 = const()[name = tensor<string, []>("attn_7_transpose_x_0"), val = tensor<bool, []>(false)]; |
| 506 | tensor<bool, []> attn_7_transpose_y_0 = const()[name = tensor<string, []>("attn_7_transpose_y_0"), val = tensor<bool, []>(true)]; |
| 507 | tensor<fp16, [1, 12, 64, 1500]> attn_7_cast_fp16 = matmul(transpose_x = attn_7_transpose_x_0, transpose_y = attn_7_transpose_y_0, x = var_851_cast_fp16, y = var_849_cast_fp16)[name = tensor<string, []>("attn_7_cast_fp16")]; |
| 508 | tensor<int32, [4]> var_854 = const()[name = tensor<string, []>("op_854"), val = tensor<int32, [4]>([1, 768, 1, 1500])]; |
| 509 | tensor<fp16, [1, 768, 1, 1500]> input_25_cast_fp16 = reshape(shape = var_854, x = attn_7_cast_fp16)[name = tensor<string, []>("input_25_cast_fp16")]; |
| 510 | tensor<string, []> var_864_pad_type_0 = const()[name = tensor<string, []>("op_864_pad_type_0"), val = tensor<string, []>("valid")]; |
| 511 | tensor<int32, [2]> var_864_strides_0 = const()[name = tensor<string, []>("op_864_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 512 | tensor<int32, [4]> var_864_pad_0 = const()[name = tensor<string, []>("op_864_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 513 | tensor<int32, [2]> var_864_dilations_0 = const()[name = tensor<string, []>("op_864_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 514 | tensor<int32, []> var_864_groups_0 = const()[name = tensor<string, []>("op_864_groups_0"), val = tensor<int32, []>(1)]; |
| 515 | tensor<fp16, [768, 768, 1, 1]> layers_3_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22272576))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22567552))), name = tensor<string, []>("layers_3_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 516 | tensor<fp16, [768]> layers_3_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22567680)))]; |
| 517 | tensor<fp16, [1, 768, 1, 1500]> var_864_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_864_dilations_0, groups = var_864_groups_0, pad = var_864_pad_0, pad_type = var_864_pad_type_0, strides = var_864_strides_0, weight = layers_3_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_25_cast_fp16)[name = tensor<string, []>("op_864_cast_fp16")]; |
| 518 | tensor<string, []> var_870_pad_type_0 = const()[name = tensor<string, []>("op_870_pad_type_0"), val = tensor<string, []>("valid")]; |
| 519 | tensor<int32, [2]> var_870_strides_0 = const()[name = tensor<string, []>("op_870_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 520 | tensor<int32, [4]> var_870_pad_0 = const()[name = tensor<string, []>("op_870_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 521 | tensor<int32, [2]> var_870_dilations_0 = const()[name = tensor<string, []>("op_870_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 522 | tensor<int32, []> var_870_groups_0 = const()[name = tensor<string, []>("op_870_groups_0"), val = tensor<int32, []>(1)]; |
| 523 | tensor<fp16, [768, 768, 1, 1]> layers_3_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22576640))), name = tensor<string, []>("layers_3_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [3641]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22569280))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 524 | tensor<fp16, [1, 768, 1, 1500]> var_870_cast_fp16 = conv(dilations = var_870_dilations_0, groups = var_870_groups_0, pad = var_870_pad_0, pad_type = var_870_pad_type_0, strides = var_870_strides_0, weight = layers_3_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_25_cast_fp16)[name = tensor<string, []>("op_870_cast_fp16")]; |
| 525 | tensor<fp16, [1, 768, 1, 1500]> obj_15_cast_fp16 = add(x = var_864_cast_fp16, y = var_870_cast_fp16)[name = tensor<string, []>("obj_15_cast_fp16")]; |
| 526 | tensor<fp16, [1, 768, 1, 1500]> inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_15_cast_fp16)[name = tensor<string, []>("inputs_15_cast_fp16")]; |
| 527 | tensor<int32, [1]> out_15_axes_0 = const()[name = tensor<string, []>("out_15_axes_0"), val = tensor<int32, [1]>([1])]; |
| 528 | tensor<fp16, []> var_881_to_fp16 = const()[name = tensor<string, []>("op_881_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 529 | tensor<fp16, [1, 768, 1, 1500]> out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_881_to_fp16, x = inputs_15_cast_fp16)[name = tensor<string, []>("out_15_cast_fp16")]; |
| 530 | tensor<fp16, [768]> input_27_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_27_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22650432)))]; |
| 531 | tensor<fp16, [768]> input_27_beta_0_to_fp16 = const()[name = tensor<string, []>("input_27_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22652032)))]; |
| 532 | tensor<fp16, []> input_27_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_27_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 533 | tensor<fp16, [1, 768, 1, 1500]> input_27_cast_fp16 = batch_norm(beta = input_27_beta_0_to_fp16, epsilon = input_27_epsilon_0_to_fp16, gamma = input_27_gamma_0_to_fp16, mean = var_57_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_15_cast_fp16)[name = tensor<string, []>("input_27_cast_fp16")]; |
| 534 | tensor<string, []> var_899_pad_type_0 = const()[name = tensor<string, []>("op_899_pad_type_0"), val = tensor<string, []>("valid")]; |
| 535 | tensor<int32, [2]> var_899_strides_0 = const()[name = tensor<string, []>("op_899_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 536 | tensor<int32, [4]> var_899_pad_0 = const()[name = tensor<string, []>("op_899_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 537 | tensor<int32, [2]> var_899_dilations_0 = const()[name = tensor<string, []>("op_899_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 538 | tensor<int32, []> var_899_groups_0 = const()[name = tensor<string, []>("op_899_groups_0"), val = tensor<int32, []>(1)]; |
| 539 | tensor<fp16, [3072, 768, 1, 1]> layers_3_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1179648]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22653632))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23833344))), name = tensor<string, []>("layers_3_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([3072, 768, 1, 1])]; |
| 540 | tensor<fp16, [3072]> layers_3_fc1_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_fc1_inlier_module_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23833472)))]; |
| 541 | tensor<fp16, [1, 3072, 1, 1500]> var_899_cast_fp16 = conv(bias = layers_3_fc1_inlier_module_bias_to_fp16, dilations = var_899_dilations_0, groups = var_899_groups_0, pad = var_899_pad_0, pad_type = var_899_pad_type_0, strides = var_899_strides_0, weight = layers_3_fc1_inlier_module_weight_to_fp16_palettized, x = input_27_cast_fp16)[name = tensor<string, []>("op_899_cast_fp16")]; |
| 542 | tensor<string, []> var_905_pad_type_0 = const()[name = tensor<string, []>("op_905_pad_type_0"), val = tensor<string, []>("valid")]; |
| 543 | tensor<int32, [2]> var_905_strides_0 = const()[name = tensor<string, []>("op_905_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 544 | tensor<int32, [4]> var_905_pad_0 = const()[name = tensor<string, []>("op_905_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 545 | tensor<int32, [2]> var_905_dilations_0 = const()[name = tensor<string, []>("op_905_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 546 | tensor<int32, []> var_905_groups_0 = const()[name = tensor<string, []>("op_905_groups_0"), val = tensor<int32, []>(1)]; |
| 547 | tensor<fp16, [3072, 768, 1, 1]> layers_3_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23883648))), name = tensor<string, []>("layers_3_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [21930]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23839680))), shape = tensor<uint32, [4]>([3072, 768, 1, 1])]; |
| 548 | tensor<fp16, [1, 3072, 1, 1500]> var_905_cast_fp16 = conv(dilations = var_905_dilations_0, groups = var_905_groups_0, pad = var_905_pad_0, pad_type = var_905_pad_type_0, strides = var_905_strides_0, weight = layers_3_fc1_outlier_module_weight_to_fp16_sparsified, x = input_27_cast_fp16)[name = tensor<string, []>("op_905_cast_fp16")]; |
| 549 | tensor<fp16, [1, 3072, 1, 1500]> input_29_cast_fp16 = add(x = var_899_cast_fp16, y = var_905_cast_fp16)[name = tensor<string, []>("input_29_cast_fp16")]; |
| 550 | tensor<string, []> input_31_mode_0 = const()[name = tensor<string, []>("input_31_mode_0"), val = tensor<string, []>("EXACT")]; |
| 551 | tensor<fp16, [1, 3072, 1, 1500]> input_31_cast_fp16 = gelu(mode = input_31_mode_0, x = input_29_cast_fp16)[name = tensor<string, []>("input_31_cast_fp16")]; |
| 552 | tensor<string, []> var_916_pad_type_0 = const()[name = tensor<string, []>("op_916_pad_type_0"), val = tensor<string, []>("valid")]; |
| 553 | tensor<int32, [2]> var_916_strides_0 = const()[name = tensor<string, []>("op_916_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 554 | tensor<int32, [4]> var_916_pad_0 = const()[name = tensor<string, []>("op_916_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 555 | tensor<int32, [2]> var_916_dilations_0 = const()[name = tensor<string, []>("op_916_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 556 | tensor<int32, []> var_916_groups_0 = const()[name = tensor<string, []>("op_916_groups_0"), val = tensor<int32, []>(1)]; |
| 557 | tensor<fp16, [768, 3072, 1, 1]> layers_3_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1179648]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24178624))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25358336))), name = tensor<string, []>("layers_3_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 3072, 1, 1])]; |
| 558 | tensor<fp16, [768]> layers_3_fc2_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_3_fc2_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25358464)))]; |
| 559 | tensor<fp16, [1, 768, 1, 1500]> var_916_cast_fp16 = conv(bias = layers_3_fc2_inlier_module_bias_to_fp16, dilations = var_916_dilations_0, groups = var_916_groups_0, pad = var_916_pad_0, pad_type = var_916_pad_type_0, strides = var_916_strides_0, weight = layers_3_fc2_inlier_module_weight_to_fp16_palettized, x = input_31_cast_fp16)[name = tensor<string, []>("op_916_cast_fp16")]; |
| 560 | tensor<string, []> var_922_pad_type_0 = const()[name = tensor<string, []>("op_922_pad_type_0"), val = tensor<string, []>("valid")]; |
| 561 | tensor<int32, [2]> var_922_strides_0 = const()[name = tensor<string, []>("op_922_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 562 | tensor<int32, [4]> var_922_pad_0 = const()[name = tensor<string, []>("op_922_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 563 | tensor<int32, [2]> var_922_dilations_0 = const()[name = tensor<string, []>("op_922_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 564 | tensor<int32, []> var_922_groups_0 = const()[name = tensor<string, []>("op_922_groups_0"), val = tensor<int32, []>(1)]; |
| 565 | tensor<fp16, [768, 3072, 1, 1]> layers_3_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25407808))), name = tensor<string, []>("layers_3_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [23820]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25360064))), shape = tensor<uint32, [4]>([768, 3072, 1, 1])]; |
| 566 | tensor<fp16, [1, 768, 1, 1500]> var_922_cast_fp16 = conv(dilations = var_922_dilations_0, groups = var_922_groups_0, pad = var_922_pad_0, pad_type = var_922_pad_type_0, strides = var_922_strides_0, weight = layers_3_fc2_outlier_module_weight_to_fp16_sparsified, x = input_31_cast_fp16)[name = tensor<string, []>("op_922_cast_fp16")]; |
| 567 | tensor<fp16, [1, 768, 1, 1500]> hidden_states_11_cast_fp16 = add(x = var_916_cast_fp16, y = var_922_cast_fp16)[name = tensor<string, []>("hidden_states_11_cast_fp16")]; |
| 568 | tensor<fp16, [1, 768, 1, 1500]> inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = hidden_states_11_cast_fp16)[name = tensor<string, []>("inputs_17_cast_fp16")]; |
| 569 | tensor<int32, []> var_928 = const()[name = tensor<string, []>("op_928"), val = tensor<int32, []>(3)]; |
| 570 | tensor<int32, [1]> out_17_axes_0 = const()[name = tensor<string, []>("out_17_axes_0"), val = tensor<int32, [1]>([1])]; |
| 571 | tensor<fp16, []> var_950_to_fp16 = const()[name = tensor<string, []>("op_950_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 572 | tensor<fp16, [1, 768, 1, 1500]> out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_950_to_fp16, x = inputs_17_cast_fp16)[name = tensor<string, []>("out_17_cast_fp16")]; |
| 573 | tensor<fp16, [768]> obj_17_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_17_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25702784)))]; |
| 574 | tensor<fp16, [768]> obj_17_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_17_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25704384)))]; |
| 575 | tensor<fp16, []> obj_17_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_17_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 576 | tensor<fp16, [1, 768, 1, 1500]> obj_17_cast_fp16 = batch_norm(beta = obj_17_beta_0_to_fp16, epsilon = obj_17_epsilon_0_to_fp16, gamma = obj_17_gamma_0_to_fp16, mean = var_57_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_17_cast_fp16)[name = tensor<string, []>("obj_17_cast_fp16")]; |
| 577 | tensor<string, []> var_972_pad_type_0 = const()[name = tensor<string, []>("op_972_pad_type_0"), val = tensor<string, []>("valid")]; |
| 578 | tensor<int32, [2]> var_972_strides_0 = const()[name = tensor<string, []>("op_972_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 579 | tensor<int32, [4]> var_972_pad_0 = const()[name = tensor<string, []>("op_972_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 580 | tensor<int32, [2]> var_972_dilations_0 = const()[name = tensor<string, []>("op_972_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 581 | tensor<int32, []> var_972_groups_0 = const()[name = tensor<string, []>("op_972_groups_0"), val = tensor<int32, []>(1)]; |
| 582 | tensor<fp16, [768, 768, 1, 1]> layers_4_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25705984))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26000960))), name = tensor<string, []>("layers_4_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 583 | tensor<fp16, [768]> layers_4_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26001088)))]; |
| 584 | tensor<fp16, [1, 768, 1, 1500]> var_972_cast_fp16 = conv(bias = layers_4_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_972_dilations_0, groups = var_972_groups_0, pad = var_972_pad_0, pad_type = var_972_pad_type_0, strides = var_972_strides_0, weight = layers_4_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_17_cast_fp16)[name = tensor<string, []>("op_972_cast_fp16")]; |
| 585 | tensor<string, []> var_978_pad_type_0 = const()[name = tensor<string, []>("op_978_pad_type_0"), val = tensor<string, []>("valid")]; |
| 586 | tensor<int32, [2]> var_978_strides_0 = const()[name = tensor<string, []>("op_978_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 587 | tensor<int32, [4]> var_978_pad_0 = const()[name = tensor<string, []>("op_978_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 588 | tensor<int32, [2]> var_978_dilations_0 = const()[name = tensor<string, []>("op_978_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 589 | tensor<int32, []> var_978_groups_0 = const()[name = tensor<string, []>("op_978_groups_0"), val = tensor<int32, []>(1)]; |
| 590 | tensor<fp16, [768, 768, 1, 1]> layers_4_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26012480))), name = tensor<string, []>("layers_4_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [4848]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26002688))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 591 | tensor<fp16, [1, 768, 1, 1500]> var_978_cast_fp16 = conv(dilations = var_978_dilations_0, groups = var_978_groups_0, pad = var_978_pad_0, pad_type = var_978_pad_type_0, strides = var_978_strides_0, weight = layers_4_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_17_cast_fp16)[name = tensor<string, []>("op_978_cast_fp16")]; |
| 592 | tensor<fp16, [1, 768, 1, 1500]> query_9_cast_fp16 = add(x = var_972_cast_fp16, y = var_978_cast_fp16)[name = tensor<string, []>("query_9_cast_fp16")]; |
| 593 | tensor<string, []> var_987_pad_type_0 = const()[name = tensor<string, []>("op_987_pad_type_0"), val = tensor<string, []>("valid")]; |
| 594 | tensor<int32, [2]> var_987_strides_0 = const()[name = tensor<string, []>("op_987_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 595 | tensor<int32, [4]> var_987_pad_0 = const()[name = tensor<string, []>("op_987_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 596 | tensor<int32, [2]> var_987_dilations_0 = const()[name = tensor<string, []>("op_987_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 597 | tensor<int32, []> var_987_groups_0 = const()[name = tensor<string, []>("op_987_groups_0"), val = tensor<int32, []>(1)]; |
| 598 | tensor<fp16, [768, 768, 1, 1]> layers_4_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26086272))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26381248))), name = tensor<string, []>("layers_4_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 599 | tensor<fp16, [1, 768, 1, 1500]> var_987_cast_fp16 = conv(dilations = var_987_dilations_0, groups = var_987_groups_0, pad = var_987_pad_0, pad_type = var_987_pad_type_0, strides = var_987_strides_0, weight = layers_4_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_17_cast_fp16)[name = tensor<string, []>("op_987_cast_fp16")]; |
| 600 | tensor<string, []> var_993_pad_type_0 = const()[name = tensor<string, []>("op_993_pad_type_0"), val = tensor<string, []>("valid")]; |
| 601 | tensor<int32, [2]> var_993_strides_0 = const()[name = tensor<string, []>("op_993_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 602 | tensor<int32, [4]> var_993_pad_0 = const()[name = tensor<string, []>("op_993_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 603 | tensor<int32, [2]> var_993_dilations_0 = const()[name = tensor<string, []>("op_993_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 604 | tensor<int32, []> var_993_groups_0 = const()[name = tensor<string, []>("op_993_groups_0"), val = tensor<int32, []>(1)]; |
| 605 | tensor<fp16, [768, 768, 1, 1]> layers_4_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26392192))), name = tensor<string, []>("layers_4_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [5362]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26381376))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 606 | tensor<fp16, [1, 768, 1, 1500]> var_993_cast_fp16 = conv(dilations = var_993_dilations_0, groups = var_993_groups_0, pad = var_993_pad_0, pad_type = var_993_pad_type_0, strides = var_993_strides_0, weight = layers_4_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_17_cast_fp16)[name = tensor<string, []>("op_993_cast_fp16")]; |
| 607 | tensor<fp16, [1, 768, 1, 1500]> key_9_cast_fp16 = add(x = var_987_cast_fp16, y = var_993_cast_fp16)[name = tensor<string, []>("key_9_cast_fp16")]; |
| 608 | tensor<string, []> var_1003_pad_type_0 = const()[name = tensor<string, []>("op_1003_pad_type_0"), val = tensor<string, []>("valid")]; |
| 609 | tensor<int32, [2]> var_1003_strides_0 = const()[name = tensor<string, []>("op_1003_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 610 | tensor<int32, [4]> var_1003_pad_0 = const()[name = tensor<string, []>("op_1003_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 611 | tensor<int32, [2]> var_1003_dilations_0 = const()[name = tensor<string, []>("op_1003_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 612 | tensor<int32, []> var_1003_groups_0 = const()[name = tensor<string, []>("op_1003_groups_0"), val = tensor<int32, []>(1)]; |
| 613 | tensor<fp16, [768, 768, 1, 1]> layers_4_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26465984))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26760960))), name = tensor<string, []>("layers_4_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 614 | tensor<fp16, [768]> layers_4_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26761088)))]; |
| 615 | tensor<fp16, [1, 768, 1, 1500]> var_1003_cast_fp16 = conv(bias = layers_4_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_1003_dilations_0, groups = var_1003_groups_0, pad = var_1003_pad_0, pad_type = var_1003_pad_type_0, strides = var_1003_strides_0, weight = layers_4_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_17_cast_fp16)[name = tensor<string, []>("op_1003_cast_fp16")]; |
| 616 | tensor<string, []> var_1009_pad_type_0 = const()[name = tensor<string, []>("op_1009_pad_type_0"), val = tensor<string, []>("valid")]; |
| 617 | tensor<int32, [2]> var_1009_strides_0 = const()[name = tensor<string, []>("op_1009_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 618 | tensor<int32, [4]> var_1009_pad_0 = const()[name = tensor<string, []>("op_1009_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 619 | tensor<int32, [2]> var_1009_dilations_0 = const()[name = tensor<string, []>("op_1009_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 620 | tensor<int32, []> var_1009_groups_0 = const()[name = tensor<string, []>("op_1009_groups_0"), val = tensor<int32, []>(1)]; |
| 621 | tensor<fp16, [768, 768, 1, 1]> layers_4_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26769984))), name = tensor<string, []>("layers_4_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [3606]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26762688))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 622 | tensor<fp16, [1, 768, 1, 1500]> var_1009_cast_fp16 = conv(dilations = var_1009_dilations_0, groups = var_1009_groups_0, pad = var_1009_pad_0, pad_type = var_1009_pad_type_0, strides = var_1009_strides_0, weight = layers_4_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_17_cast_fp16)[name = tensor<string, []>("op_1009_cast_fp16")]; |
| 623 | tensor<fp16, [1, 768, 1, 1500]> value_9_cast_fp16 = add(x = var_1003_cast_fp16, y = var_1009_cast_fp16)[name = tensor<string, []>("value_9_cast_fp16")]; |
| 624 | tensor<int32, [4]> var_1013 = const()[name = tensor<string, []>("op_1013"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 625 | tensor<fp16, [1, 12, 64, 1500]> mh_q_9_cast_fp16 = reshape(shape = var_1013, x = query_9_cast_fp16)[name = tensor<string, []>("mh_q_9_cast_fp16")]; |
| 626 | tensor<fp16, []> var_1015_to_fp16 = const()[name = tensor<string, []>("op_1015_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| 627 | tensor<fp16, [1, 12, 64, 1500]> var_1016_cast_fp16 = mul(x = mh_q_9_cast_fp16, y = var_1015_to_fp16)[name = tensor<string, []>("op_1016_cast_fp16")]; |
| 628 | tensor<int32, [4]> var_1019 = const()[name = tensor<string, []>("op_1019"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 629 | tensor<fp16, [1, 12, 64, 1500]> var_1020_cast_fp16 = reshape(shape = var_1019, x = key_9_cast_fp16)[name = tensor<string, []>("op_1020_cast_fp16")]; |
| 630 | tensor<bool, []> mh_w_9_transpose_x_0 = const()[name = tensor<string, []>("mh_w_9_transpose_x_0"), val = tensor<bool, []>(true)]; |
| 631 | tensor<bool, []> mh_w_9_transpose_y_0 = const()[name = tensor<string, []>("mh_w_9_transpose_y_0"), val = tensor<bool, []>(false)]; |
| 632 | tensor<fp16, [1, 12, 1500, 1500]> mh_w_9_cast_fp16 = matmul(transpose_x = mh_w_9_transpose_x_0, transpose_y = mh_w_9_transpose_y_0, x = var_1016_cast_fp16, y = var_1020_cast_fp16)[name = tensor<string, []>("mh_w_9_cast_fp16")]; |
| 633 | tensor<fp16, [1, 12, 1500, 1500]> var_1023_cast_fp16 = softmax(axis = var_928, x = mh_w_9_cast_fp16)[name = tensor<string, []>("op_1023_cast_fp16")]; |
| 634 | tensor<int32, [4]> var_1024 = const()[name = tensor<string, []>("op_1024"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 635 | tensor<fp16, [1, 12, 64, 1500]> var_1025_cast_fp16 = reshape(shape = var_1024, x = value_9_cast_fp16)[name = tensor<string, []>("op_1025_cast_fp16")]; |
| 636 | tensor<bool, []> attn_9_transpose_x_0 = const()[name = tensor<string, []>("attn_9_transpose_x_0"), val = tensor<bool, []>(false)]; |
| 637 | tensor<bool, []> attn_9_transpose_y_0 = const()[name = tensor<string, []>("attn_9_transpose_y_0"), val = tensor<bool, []>(true)]; |
| 638 | tensor<fp16, [1, 12, 64, 1500]> attn_9_cast_fp16 = matmul(transpose_x = attn_9_transpose_x_0, transpose_y = attn_9_transpose_y_0, x = var_1025_cast_fp16, y = var_1023_cast_fp16)[name = tensor<string, []>("attn_9_cast_fp16")]; |
| 639 | tensor<int32, [4]> var_1028 = const()[name = tensor<string, []>("op_1028"), val = tensor<int32, [4]>([1, 768, 1, 1500])]; |
| 640 | tensor<fp16, [1, 768, 1, 1500]> input_33_cast_fp16 = reshape(shape = var_1028, x = attn_9_cast_fp16)[name = tensor<string, []>("input_33_cast_fp16")]; |
| 641 | tensor<string, []> var_1038_pad_type_0 = const()[name = tensor<string, []>("op_1038_pad_type_0"), val = tensor<string, []>("valid")]; |
| 642 | tensor<int32, [2]> var_1038_strides_0 = const()[name = tensor<string, []>("op_1038_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 643 | tensor<int32, [4]> var_1038_pad_0 = const()[name = tensor<string, []>("op_1038_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 644 | tensor<int32, [2]> var_1038_dilations_0 = const()[name = tensor<string, []>("op_1038_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 645 | tensor<int32, []> var_1038_groups_0 = const()[name = tensor<string, []>("op_1038_groups_0"), val = tensor<int32, []>(1)]; |
| 646 | tensor<fp16, [768, 768, 1, 1]> layers_4_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26843776))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27138752))), name = tensor<string, []>("layers_4_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 647 | tensor<fp16, [768]> layers_4_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27138880)))]; |
| 648 | tensor<fp16, [1, 768, 1, 1500]> var_1038_cast_fp16 = conv(bias = layers_4_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_1038_dilations_0, groups = var_1038_groups_0, pad = var_1038_pad_0, pad_type = var_1038_pad_type_0, strides = var_1038_strides_0, weight = layers_4_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_33_cast_fp16)[name = tensor<string, []>("op_1038_cast_fp16")]; |
| 649 | tensor<string, []> var_1044_pad_type_0 = const()[name = tensor<string, []>("op_1044_pad_type_0"), val = tensor<string, []>("valid")]; |
| 650 | tensor<int32, [2]> var_1044_strides_0 = const()[name = tensor<string, []>("op_1044_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 651 | tensor<int32, [4]> var_1044_pad_0 = const()[name = tensor<string, []>("op_1044_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 652 | tensor<int32, [2]> var_1044_dilations_0 = const()[name = tensor<string, []>("op_1044_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 653 | tensor<int32, []> var_1044_groups_0 = const()[name = tensor<string, []>("op_1044_groups_0"), val = tensor<int32, []>(1)]; |
| 654 | tensor<fp16, [768, 768, 1, 1]> layers_4_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27146816))), name = tensor<string, []>("layers_4_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [3133]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27140480))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 655 | tensor<fp16, [1, 768, 1, 1500]> var_1044_cast_fp16 = conv(dilations = var_1044_dilations_0, groups = var_1044_groups_0, pad = var_1044_pad_0, pad_type = var_1044_pad_type_0, strides = var_1044_strides_0, weight = layers_4_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_33_cast_fp16)[name = tensor<string, []>("op_1044_cast_fp16")]; |
| 656 | tensor<fp16, [1, 768, 1, 1500]> obj_19_cast_fp16 = add(x = var_1038_cast_fp16, y = var_1044_cast_fp16)[name = tensor<string, []>("obj_19_cast_fp16")]; |
| 657 | tensor<fp16, [1, 768, 1, 1500]> inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = obj_19_cast_fp16)[name = tensor<string, []>("inputs_19_cast_fp16")]; |
| 658 | tensor<int32, [1]> out_19_axes_0 = const()[name = tensor<string, []>("out_19_axes_0"), val = tensor<int32, [1]>([1])]; |
| 659 | tensor<fp16, []> var_1055_to_fp16 = const()[name = tensor<string, []>("op_1055_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 660 | tensor<fp16, [1, 768, 1, 1500]> out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_1055_to_fp16, x = inputs_19_cast_fp16)[name = tensor<string, []>("out_19_cast_fp16")]; |
| 661 | tensor<fp16, [768]> input_35_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_35_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27220608)))]; |
| 662 | tensor<fp16, [768]> input_35_beta_0_to_fp16 = const()[name = tensor<string, []>("input_35_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27222208)))]; |
| 663 | tensor<fp16, []> input_35_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_35_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 664 | tensor<fp16, [1, 768, 1, 1500]> input_35_cast_fp16 = batch_norm(beta = input_35_beta_0_to_fp16, epsilon = input_35_epsilon_0_to_fp16, gamma = input_35_gamma_0_to_fp16, mean = var_57_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_19_cast_fp16)[name = tensor<string, []>("input_35_cast_fp16")]; |
| 665 | tensor<string, []> var_1073_pad_type_0 = const()[name = tensor<string, []>("op_1073_pad_type_0"), val = tensor<string, []>("valid")]; |
| 666 | tensor<int32, [2]> var_1073_strides_0 = const()[name = tensor<string, []>("op_1073_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 667 | tensor<int32, [4]> var_1073_pad_0 = const()[name = tensor<string, []>("op_1073_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 668 | tensor<int32, [2]> var_1073_dilations_0 = const()[name = tensor<string, []>("op_1073_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 669 | tensor<int32, []> var_1073_groups_0 = const()[name = tensor<string, []>("op_1073_groups_0"), val = tensor<int32, []>(1)]; |
| 670 | tensor<fp16, [3072, 768, 1, 1]> layers_4_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1179648]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27223808))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28403520))), name = tensor<string, []>("layers_4_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([3072, 768, 1, 1])]; |
| 671 | tensor<fp16, [3072]> layers_4_fc1_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_fc1_inlier_module_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28403648)))]; |
| 672 | tensor<fp16, [1, 3072, 1, 1500]> var_1073_cast_fp16 = conv(bias = layers_4_fc1_inlier_module_bias_to_fp16, dilations = var_1073_dilations_0, groups = var_1073_groups_0, pad = var_1073_pad_0, pad_type = var_1073_pad_type_0, strides = var_1073_strides_0, weight = layers_4_fc1_inlier_module_weight_to_fp16_palettized, x = input_35_cast_fp16)[name = tensor<string, []>("op_1073_cast_fp16")]; |
| 673 | tensor<string, []> var_1079_pad_type_0 = const()[name = tensor<string, []>("op_1079_pad_type_0"), val = tensor<string, []>("valid")]; |
| 674 | tensor<int32, [2]> var_1079_strides_0 = const()[name = tensor<string, []>("op_1079_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 675 | tensor<int32, [4]> var_1079_pad_0 = const()[name = tensor<string, []>("op_1079_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 676 | tensor<int32, [2]> var_1079_dilations_0 = const()[name = tensor<string, []>("op_1079_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 677 | tensor<int32, []> var_1079_groups_0 = const()[name = tensor<string, []>("op_1079_groups_0"), val = tensor<int32, []>(1)]; |
| 678 | tensor<fp16, [3072, 768, 1, 1]> layers_4_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28452288))), name = tensor<string, []>("layers_4_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [21163]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28409856))), shape = tensor<uint32, [4]>([3072, 768, 1, 1])]; |
| 679 | tensor<fp16, [1, 3072, 1, 1500]> var_1079_cast_fp16 = conv(dilations = var_1079_dilations_0, groups = var_1079_groups_0, pad = var_1079_pad_0, pad_type = var_1079_pad_type_0, strides = var_1079_strides_0, weight = layers_4_fc1_outlier_module_weight_to_fp16_sparsified, x = input_35_cast_fp16)[name = tensor<string, []>("op_1079_cast_fp16")]; |
| 680 | tensor<fp16, [1, 3072, 1, 1500]> input_37_cast_fp16 = add(x = var_1073_cast_fp16, y = var_1079_cast_fp16)[name = tensor<string, []>("input_37_cast_fp16")]; |
| 681 | tensor<string, []> input_39_mode_0 = const()[name = tensor<string, []>("input_39_mode_0"), val = tensor<string, []>("EXACT")]; |
| 682 | tensor<fp16, [1, 3072, 1, 1500]> input_39_cast_fp16 = gelu(mode = input_39_mode_0, x = input_37_cast_fp16)[name = tensor<string, []>("input_39_cast_fp16")]; |
| 683 | tensor<string, []> var_1090_pad_type_0 = const()[name = tensor<string, []>("op_1090_pad_type_0"), val = tensor<string, []>("valid")]; |
| 684 | tensor<int32, [2]> var_1090_strides_0 = const()[name = tensor<string, []>("op_1090_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 685 | tensor<int32, [4]> var_1090_pad_0 = const()[name = tensor<string, []>("op_1090_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 686 | tensor<int32, [2]> var_1090_dilations_0 = const()[name = tensor<string, []>("op_1090_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 687 | tensor<int32, []> var_1090_groups_0 = const()[name = tensor<string, []>("op_1090_groups_0"), val = tensor<int32, []>(1)]; |
| 688 | tensor<fp16, [768, 3072, 1, 1]> layers_4_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1179648]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28747264))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29926976))), name = tensor<string, []>("layers_4_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 3072, 1, 1])]; |
| 689 | tensor<fp16, [768]> layers_4_fc2_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_4_fc2_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29927104)))]; |
| 690 | tensor<fp16, [1, 768, 1, 1500]> var_1090_cast_fp16 = conv(bias = layers_4_fc2_inlier_module_bias_to_fp16, dilations = var_1090_dilations_0, groups = var_1090_groups_0, pad = var_1090_pad_0, pad_type = var_1090_pad_type_0, strides = var_1090_strides_0, weight = layers_4_fc2_inlier_module_weight_to_fp16_palettized, x = input_39_cast_fp16)[name = tensor<string, []>("op_1090_cast_fp16")]; |
| 691 | tensor<string, []> var_1096_pad_type_0 = const()[name = tensor<string, []>("op_1096_pad_type_0"), val = tensor<string, []>("valid")]; |
| 692 | tensor<int32, [2]> var_1096_strides_0 = const()[name = tensor<string, []>("op_1096_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 693 | tensor<int32, [4]> var_1096_pad_0 = const()[name = tensor<string, []>("op_1096_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 694 | tensor<int32, [2]> var_1096_dilations_0 = const()[name = tensor<string, []>("op_1096_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 695 | tensor<int32, []> var_1096_groups_0 = const()[name = tensor<string, []>("op_1096_groups_0"), val = tensor<int32, []>(1)]; |
| 696 | tensor<fp16, [768, 3072, 1, 1]> layers_4_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29977152))), name = tensor<string, []>("layers_4_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [24181]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29928704))), shape = tensor<uint32, [4]>([768, 3072, 1, 1])]; |
| 697 | tensor<fp16, [1, 768, 1, 1500]> var_1096_cast_fp16 = conv(dilations = var_1096_dilations_0, groups = var_1096_groups_0, pad = var_1096_pad_0, pad_type = var_1096_pad_type_0, strides = var_1096_strides_0, weight = layers_4_fc2_outlier_module_weight_to_fp16_sparsified, x = input_39_cast_fp16)[name = tensor<string, []>("op_1096_cast_fp16")]; |
| 698 | tensor<fp16, [1, 768, 1, 1500]> hidden_states_13_cast_fp16 = add(x = var_1090_cast_fp16, y = var_1096_cast_fp16)[name = tensor<string, []>("hidden_states_13_cast_fp16")]; |
| 699 | tensor<fp16, [1, 768, 1, 1500]> inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = hidden_states_13_cast_fp16)[name = tensor<string, []>("inputs_21_cast_fp16")]; |
| 700 | tensor<int32, []> var_1102 = const()[name = tensor<string, []>("op_1102"), val = tensor<int32, []>(3)]; |
| 701 | tensor<int32, [1]> out_21_axes_0 = const()[name = tensor<string, []>("out_21_axes_0"), val = tensor<int32, [1]>([1])]; |
| 702 | tensor<fp16, []> var_1124_to_fp16 = const()[name = tensor<string, []>("op_1124_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 703 | tensor<fp16, [1, 768, 1, 1500]> out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_1124_to_fp16, x = inputs_21_cast_fp16)[name = tensor<string, []>("out_21_cast_fp16")]; |
| 704 | tensor<fp16, [768]> obj_21_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_21_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30272128)))]; |
| 705 | tensor<fp16, [768]> obj_21_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_21_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30273728)))]; |
| 706 | tensor<fp16, []> obj_21_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_21_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 707 | tensor<fp16, [1, 768, 1, 1500]> obj_21_cast_fp16 = batch_norm(beta = obj_21_beta_0_to_fp16, epsilon = obj_21_epsilon_0_to_fp16, gamma = obj_21_gamma_0_to_fp16, mean = var_57_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_21_cast_fp16)[name = tensor<string, []>("obj_21_cast_fp16")]; |
| 708 | tensor<string, []> var_1146_pad_type_0 = const()[name = tensor<string, []>("op_1146_pad_type_0"), val = tensor<string, []>("valid")]; |
| 709 | tensor<int32, [2]> var_1146_strides_0 = const()[name = tensor<string, []>("op_1146_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 710 | tensor<int32, [4]> var_1146_pad_0 = const()[name = tensor<string, []>("op_1146_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 711 | tensor<int32, [2]> var_1146_dilations_0 = const()[name = tensor<string, []>("op_1146_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 712 | tensor<int32, []> var_1146_groups_0 = const()[name = tensor<string, []>("op_1146_groups_0"), val = tensor<int32, []>(1)]; |
| 713 | tensor<fp16, [768, 768, 1, 1]> layers_5_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30275328))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30570304))), name = tensor<string, []>("layers_5_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 714 | tensor<fp16, [768]> layers_5_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30570432)))]; |
| 715 | tensor<fp16, [1, 768, 1, 1500]> var_1146_cast_fp16 = conv(bias = layers_5_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_1146_dilations_0, groups = var_1146_groups_0, pad = var_1146_pad_0, pad_type = var_1146_pad_type_0, strides = var_1146_strides_0, weight = layers_5_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_21_cast_fp16)[name = tensor<string, []>("op_1146_cast_fp16")]; |
| 716 | tensor<string, []> var_1152_pad_type_0 = const()[name = tensor<string, []>("op_1152_pad_type_0"), val = tensor<string, []>("valid")]; |
| 717 | tensor<int32, [2]> var_1152_strides_0 = const()[name = tensor<string, []>("op_1152_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 718 | tensor<int32, [4]> var_1152_pad_0 = const()[name = tensor<string, []>("op_1152_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 719 | tensor<int32, [2]> var_1152_dilations_0 = const()[name = tensor<string, []>("op_1152_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 720 | tensor<int32, []> var_1152_groups_0 = const()[name = tensor<string, []>("op_1152_groups_0"), val = tensor<int32, []>(1)]; |
| 721 | tensor<fp16, [768, 768, 1, 1]> layers_5_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30579968))), name = tensor<string, []>("layers_5_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [3910]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30572032))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 722 | tensor<fp16, [1, 768, 1, 1500]> var_1152_cast_fp16 = conv(dilations = var_1152_dilations_0, groups = var_1152_groups_0, pad = var_1152_pad_0, pad_type = var_1152_pad_type_0, strides = var_1152_strides_0, weight = layers_5_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_21_cast_fp16)[name = tensor<string, []>("op_1152_cast_fp16")]; |
| 723 | tensor<fp16, [1, 768, 1, 1500]> query_11_cast_fp16 = add(x = var_1146_cast_fp16, y = var_1152_cast_fp16)[name = tensor<string, []>("query_11_cast_fp16")]; |
| 724 | tensor<string, []> var_1161_pad_type_0 = const()[name = tensor<string, []>("op_1161_pad_type_0"), val = tensor<string, []>("valid")]; |
| 725 | tensor<int32, [2]> var_1161_strides_0 = const()[name = tensor<string, []>("op_1161_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 726 | tensor<int32, [4]> var_1161_pad_0 = const()[name = tensor<string, []>("op_1161_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 727 | tensor<int32, [2]> var_1161_dilations_0 = const()[name = tensor<string, []>("op_1161_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 728 | tensor<int32, []> var_1161_groups_0 = const()[name = tensor<string, []>("op_1161_groups_0"), val = tensor<int32, []>(1)]; |
| 729 | tensor<fp16, [768, 768, 1, 1]> layers_5_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30653760))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30948736))), name = tensor<string, []>("layers_5_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 730 | tensor<fp16, [1, 768, 1, 1500]> var_1161_cast_fp16 = conv(dilations = var_1161_dilations_0, groups = var_1161_groups_0, pad = var_1161_pad_0, pad_type = var_1161_pad_type_0, strides = var_1161_strides_0, weight = layers_5_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_21_cast_fp16)[name = tensor<string, []>("op_1161_cast_fp16")]; |
| 731 | tensor<string, []> var_1167_pad_type_0 = const()[name = tensor<string, []>("op_1167_pad_type_0"), val = tensor<string, []>("valid")]; |
| 732 | tensor<int32, [2]> var_1167_strides_0 = const()[name = tensor<string, []>("op_1167_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 733 | tensor<int32, [4]> var_1167_pad_0 = const()[name = tensor<string, []>("op_1167_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 734 | tensor<int32, [2]> var_1167_dilations_0 = const()[name = tensor<string, []>("op_1167_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 735 | tensor<int32, []> var_1167_groups_0 = const()[name = tensor<string, []>("op_1167_groups_0"), val = tensor<int32, []>(1)]; |
| 736 | tensor<fp16, [768, 768, 1, 1]> layers_5_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30957056))), name = tensor<string, []>("layers_5_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [4049]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30948864))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 737 | tensor<fp16, [1, 768, 1, 1500]> var_1167_cast_fp16 = conv(dilations = var_1167_dilations_0, groups = var_1167_groups_0, pad = var_1167_pad_0, pad_type = var_1167_pad_type_0, strides = var_1167_strides_0, weight = layers_5_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_21_cast_fp16)[name = tensor<string, []>("op_1167_cast_fp16")]; |
| 738 | tensor<fp16, [1, 768, 1, 1500]> key_11_cast_fp16 = add(x = var_1161_cast_fp16, y = var_1167_cast_fp16)[name = tensor<string, []>("key_11_cast_fp16")]; |
| 739 | tensor<string, []> var_1177_pad_type_0 = const()[name = tensor<string, []>("op_1177_pad_type_0"), val = tensor<string, []>("valid")]; |
| 740 | tensor<int32, [2]> var_1177_strides_0 = const()[name = tensor<string, []>("op_1177_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 741 | tensor<int32, [4]> var_1177_pad_0 = const()[name = tensor<string, []>("op_1177_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 742 | tensor<int32, [2]> var_1177_dilations_0 = const()[name = tensor<string, []>("op_1177_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 743 | tensor<int32, []> var_1177_groups_0 = const()[name = tensor<string, []>("op_1177_groups_0"), val = tensor<int32, []>(1)]; |
| 744 | tensor<fp16, [768, 768, 1, 1]> layers_5_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31030848))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31325824))), name = tensor<string, []>("layers_5_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 745 | tensor<fp16, [768]> layers_5_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31325952)))]; |
| 746 | tensor<fp16, [1, 768, 1, 1500]> var_1177_cast_fp16 = conv(bias = layers_5_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_1177_dilations_0, groups = var_1177_groups_0, pad = var_1177_pad_0, pad_type = var_1177_pad_type_0, strides = var_1177_strides_0, weight = layers_5_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_21_cast_fp16)[name = tensor<string, []>("op_1177_cast_fp16")]; |
| 747 | tensor<string, []> var_1183_pad_type_0 = const()[name = tensor<string, []>("op_1183_pad_type_0"), val = tensor<string, []>("valid")]; |
| 748 | tensor<int32, [2]> var_1183_strides_0 = const()[name = tensor<string, []>("op_1183_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 749 | tensor<int32, [4]> var_1183_pad_0 = const()[name = tensor<string, []>("op_1183_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 750 | tensor<int32, [2]> var_1183_dilations_0 = const()[name = tensor<string, []>("op_1183_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 751 | tensor<int32, []> var_1183_groups_0 = const()[name = tensor<string, []>("op_1183_groups_0"), val = tensor<int32, []>(1)]; |
| 752 | tensor<fp16, [768, 768, 1, 1]> layers_5_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31334976))), name = tensor<string, []>("layers_5_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [3661]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31327552))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 753 | tensor<fp16, [1, 768, 1, 1500]> var_1183_cast_fp16 = conv(dilations = var_1183_dilations_0, groups = var_1183_groups_0, pad = var_1183_pad_0, pad_type = var_1183_pad_type_0, strides = var_1183_strides_0, weight = layers_5_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_21_cast_fp16)[name = tensor<string, []>("op_1183_cast_fp16")]; |
| 754 | tensor<fp16, [1, 768, 1, 1500]> value_11_cast_fp16 = add(x = var_1177_cast_fp16, y = var_1183_cast_fp16)[name = tensor<string, []>("value_11_cast_fp16")]; |
| 755 | tensor<int32, [4]> var_1187 = const()[name = tensor<string, []>("op_1187"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 756 | tensor<fp16, [1, 12, 64, 1500]> mh_q_11_cast_fp16 = reshape(shape = var_1187, x = query_11_cast_fp16)[name = tensor<string, []>("mh_q_11_cast_fp16")]; |
| 757 | tensor<fp16, []> var_1189_to_fp16 = const()[name = tensor<string, []>("op_1189_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| 758 | tensor<fp16, [1, 12, 64, 1500]> var_1190_cast_fp16 = mul(x = mh_q_11_cast_fp16, y = var_1189_to_fp16)[name = tensor<string, []>("op_1190_cast_fp16")]; |
| 759 | tensor<int32, [4]> var_1193 = const()[name = tensor<string, []>("op_1193"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 760 | tensor<fp16, [1, 12, 64, 1500]> var_1194_cast_fp16 = reshape(shape = var_1193, x = key_11_cast_fp16)[name = tensor<string, []>("op_1194_cast_fp16")]; |
| 761 | tensor<bool, []> mh_w_11_transpose_x_0 = const()[name = tensor<string, []>("mh_w_11_transpose_x_0"), val = tensor<bool, []>(true)]; |
| 762 | tensor<bool, []> mh_w_11_transpose_y_0 = const()[name = tensor<string, []>("mh_w_11_transpose_y_0"), val = tensor<bool, []>(false)]; |
| 763 | tensor<fp16, [1, 12, 1500, 1500]> mh_w_11_cast_fp16 = matmul(transpose_x = mh_w_11_transpose_x_0, transpose_y = mh_w_11_transpose_y_0, x = var_1190_cast_fp16, y = var_1194_cast_fp16)[name = tensor<string, []>("mh_w_11_cast_fp16")]; |
| 764 | tensor<fp16, [1, 12, 1500, 1500]> var_1197_cast_fp16 = softmax(axis = var_1102, x = mh_w_11_cast_fp16)[name = tensor<string, []>("op_1197_cast_fp16")]; |
| 765 | tensor<int32, [4]> var_1198 = const()[name = tensor<string, []>("op_1198"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 766 | tensor<fp16, [1, 12, 64, 1500]> var_1199_cast_fp16 = reshape(shape = var_1198, x = value_11_cast_fp16)[name = tensor<string, []>("op_1199_cast_fp16")]; |
| 767 | tensor<bool, []> attn_11_transpose_x_0 = const()[name = tensor<string, []>("attn_11_transpose_x_0"), val = tensor<bool, []>(false)]; |
| 768 | tensor<bool, []> attn_11_transpose_y_0 = const()[name = tensor<string, []>("attn_11_transpose_y_0"), val = tensor<bool, []>(true)]; |
| 769 | tensor<fp16, [1, 12, 64, 1500]> attn_11_cast_fp16 = matmul(transpose_x = attn_11_transpose_x_0, transpose_y = attn_11_transpose_y_0, x = var_1199_cast_fp16, y = var_1197_cast_fp16)[name = tensor<string, []>("attn_11_cast_fp16")]; |
| 770 | tensor<int32, [4]> var_1202 = const()[name = tensor<string, []>("op_1202"), val = tensor<int32, [4]>([1, 768, 1, 1500])]; |
| 771 | tensor<fp16, [1, 768, 1, 1500]> input_41_cast_fp16 = reshape(shape = var_1202, x = attn_11_cast_fp16)[name = tensor<string, []>("input_41_cast_fp16")]; |
| 772 | tensor<string, []> var_1212_pad_type_0 = const()[name = tensor<string, []>("op_1212_pad_type_0"), val = tensor<string, []>("valid")]; |
| 773 | tensor<int32, [2]> var_1212_strides_0 = const()[name = tensor<string, []>("op_1212_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 774 | tensor<int32, [4]> var_1212_pad_0 = const()[name = tensor<string, []>("op_1212_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 775 | tensor<int32, [2]> var_1212_dilations_0 = const()[name = tensor<string, []>("op_1212_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 776 | tensor<int32, []> var_1212_groups_0 = const()[name = tensor<string, []>("op_1212_groups_0"), val = tensor<int32, []>(1)]; |
| 777 | tensor<fp16, [768, 768, 1, 1]> layers_5_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31408768))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31703744))), name = tensor<string, []>("layers_5_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 778 | tensor<fp16, [768]> layers_5_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31703872)))]; |
| 779 | tensor<fp16, [1, 768, 1, 1500]> var_1212_cast_fp16 = conv(bias = layers_5_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_1212_dilations_0, groups = var_1212_groups_0, pad = var_1212_pad_0, pad_type = var_1212_pad_type_0, strides = var_1212_strides_0, weight = layers_5_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_41_cast_fp16)[name = tensor<string, []>("op_1212_cast_fp16")]; |
| 780 | tensor<string, []> var_1218_pad_type_0 = const()[name = tensor<string, []>("op_1218_pad_type_0"), val = tensor<string, []>("valid")]; |
| 781 | tensor<int32, [2]> var_1218_strides_0 = const()[name = tensor<string, []>("op_1218_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 782 | tensor<int32, [4]> var_1218_pad_0 = const()[name = tensor<string, []>("op_1218_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 783 | tensor<int32, [2]> var_1218_dilations_0 = const()[name = tensor<string, []>("op_1218_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 784 | tensor<int32, []> var_1218_groups_0 = const()[name = tensor<string, []>("op_1218_groups_0"), val = tensor<int32, []>(1)]; |
| 785 | tensor<fp16, [768, 768, 1, 1]> layers_5_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31713792))), name = tensor<string, []>("layers_5_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [4128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31705472))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 786 | tensor<fp16, [1, 768, 1, 1500]> var_1218_cast_fp16 = conv(dilations = var_1218_dilations_0, groups = var_1218_groups_0, pad = var_1218_pad_0, pad_type = var_1218_pad_type_0, strides = var_1218_strides_0, weight = layers_5_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_41_cast_fp16)[name = tensor<string, []>("op_1218_cast_fp16")]; |
| 787 | tensor<fp16, [1, 768, 1, 1500]> obj_23_cast_fp16 = add(x = var_1212_cast_fp16, y = var_1218_cast_fp16)[name = tensor<string, []>("obj_23_cast_fp16")]; |
| 788 | tensor<fp16, [1, 768, 1, 1500]> inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = obj_23_cast_fp16)[name = tensor<string, []>("inputs_23_cast_fp16")]; |
| 789 | tensor<int32, [1]> out_23_axes_0 = const()[name = tensor<string, []>("out_23_axes_0"), val = tensor<int32, [1]>([1])]; |
| 790 | tensor<fp16, []> var_1229_to_fp16 = const()[name = tensor<string, []>("op_1229_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 791 | tensor<fp16, [1, 768, 1, 1500]> out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_1229_to_fp16, x = inputs_23_cast_fp16)[name = tensor<string, []>("out_23_cast_fp16")]; |
| 792 | tensor<fp16, [768]> input_43_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_43_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31787584)))]; |
| 793 | tensor<fp16, [768]> input_43_beta_0_to_fp16 = const()[name = tensor<string, []>("input_43_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31789184)))]; |
| 794 | tensor<fp16, []> input_43_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_43_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 795 | tensor<fp16, [1, 768, 1, 1500]> input_43_cast_fp16 = batch_norm(beta = input_43_beta_0_to_fp16, epsilon = input_43_epsilon_0_to_fp16, gamma = input_43_gamma_0_to_fp16, mean = var_57_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_23_cast_fp16)[name = tensor<string, []>("input_43_cast_fp16")]; |
| 796 | tensor<string, []> var_1247_pad_type_0 = const()[name = tensor<string, []>("op_1247_pad_type_0"), val = tensor<string, []>("valid")]; |
| 797 | tensor<int32, [2]> var_1247_strides_0 = const()[name = tensor<string, []>("op_1247_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 798 | tensor<int32, [4]> var_1247_pad_0 = const()[name = tensor<string, []>("op_1247_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 799 | tensor<int32, [2]> var_1247_dilations_0 = const()[name = tensor<string, []>("op_1247_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 800 | tensor<int32, []> var_1247_groups_0 = const()[name = tensor<string, []>("op_1247_groups_0"), val = tensor<int32, []>(1)]; |
| 801 | tensor<fp16, [3072, 768, 1, 1]> layers_5_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1179648]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31790784))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32970496))), name = tensor<string, []>("layers_5_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([3072, 768, 1, 1])]; |
| 802 | tensor<fp16, [3072]> layers_5_fc1_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_fc1_inlier_module_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32970624)))]; |
| 803 | tensor<fp16, [1, 3072, 1, 1500]> var_1247_cast_fp16 = conv(bias = layers_5_fc1_inlier_module_bias_to_fp16, dilations = var_1247_dilations_0, groups = var_1247_groups_0, pad = var_1247_pad_0, pad_type = var_1247_pad_type_0, strides = var_1247_strides_0, weight = layers_5_fc1_inlier_module_weight_to_fp16_palettized, x = input_43_cast_fp16)[name = tensor<string, []>("op_1247_cast_fp16")]; |
| 804 | tensor<string, []> var_1253_pad_type_0 = const()[name = tensor<string, []>("op_1253_pad_type_0"), val = tensor<string, []>("valid")]; |
| 805 | tensor<int32, [2]> var_1253_strides_0 = const()[name = tensor<string, []>("op_1253_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 806 | tensor<int32, [4]> var_1253_pad_0 = const()[name = tensor<string, []>("op_1253_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 807 | tensor<int32, [2]> var_1253_dilations_0 = const()[name = tensor<string, []>("op_1253_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 808 | tensor<int32, []> var_1253_groups_0 = const()[name = tensor<string, []>("op_1253_groups_0"), val = tensor<int32, []>(1)]; |
| 809 | tensor<fp16, [3072, 768, 1, 1]> layers_5_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33018432))), name = tensor<string, []>("layers_5_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [20754]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32976832))), shape = tensor<uint32, [4]>([3072, 768, 1, 1])]; |
| 810 | tensor<fp16, [1, 3072, 1, 1500]> var_1253_cast_fp16 = conv(dilations = var_1253_dilations_0, groups = var_1253_groups_0, pad = var_1253_pad_0, pad_type = var_1253_pad_type_0, strides = var_1253_strides_0, weight = layers_5_fc1_outlier_module_weight_to_fp16_sparsified, x = input_43_cast_fp16)[name = tensor<string, []>("op_1253_cast_fp16")]; |
| 811 | tensor<fp16, [1, 3072, 1, 1500]> input_45_cast_fp16 = add(x = var_1247_cast_fp16, y = var_1253_cast_fp16)[name = tensor<string, []>("input_45_cast_fp16")]; |
| 812 | tensor<string, []> input_47_mode_0 = const()[name = tensor<string, []>("input_47_mode_0"), val = tensor<string, []>("EXACT")]; |
| 813 | tensor<fp16, [1, 3072, 1, 1500]> input_47_cast_fp16 = gelu(mode = input_47_mode_0, x = input_45_cast_fp16)[name = tensor<string, []>("input_47_cast_fp16")]; |
| 814 | tensor<string, []> var_1264_pad_type_0 = const()[name = tensor<string, []>("op_1264_pad_type_0"), val = tensor<string, []>("valid")]; |
| 815 | tensor<int32, [2]> var_1264_strides_0 = const()[name = tensor<string, []>("op_1264_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 816 | tensor<int32, [4]> var_1264_pad_0 = const()[name = tensor<string, []>("op_1264_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 817 | tensor<int32, [2]> var_1264_dilations_0 = const()[name = tensor<string, []>("op_1264_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 818 | tensor<int32, []> var_1264_groups_0 = const()[name = tensor<string, []>("op_1264_groups_0"), val = tensor<int32, []>(1)]; |
| 819 | tensor<fp16, [768, 3072, 1, 1]> layers_5_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1179648]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33313408))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34493120))), name = tensor<string, []>("layers_5_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 3072, 1, 1])]; |
| 820 | tensor<fp16, [768]> layers_5_fc2_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_5_fc2_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34493248)))]; |
| 821 | tensor<fp16, [1, 768, 1, 1500]> var_1264_cast_fp16 = conv(bias = layers_5_fc2_inlier_module_bias_to_fp16, dilations = var_1264_dilations_0, groups = var_1264_groups_0, pad = var_1264_pad_0, pad_type = var_1264_pad_type_0, strides = var_1264_strides_0, weight = layers_5_fc2_inlier_module_weight_to_fp16_palettized, x = input_47_cast_fp16)[name = tensor<string, []>("op_1264_cast_fp16")]; |
| 822 | tensor<string, []> var_1270_pad_type_0 = const()[name = tensor<string, []>("op_1270_pad_type_0"), val = tensor<string, []>("valid")]; |
| 823 | tensor<int32, [2]> var_1270_strides_0 = const()[name = tensor<string, []>("op_1270_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 824 | tensor<int32, [4]> var_1270_pad_0 = const()[name = tensor<string, []>("op_1270_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 825 | tensor<int32, [2]> var_1270_dilations_0 = const()[name = tensor<string, []>("op_1270_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 826 | tensor<int32, []> var_1270_groups_0 = const()[name = tensor<string, []>("op_1270_groups_0"), val = tensor<int32, []>(1)]; |
| 827 | tensor<fp16, [768, 3072, 1, 1]> layers_5_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34539520))), name = tensor<string, []>("layers_5_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [22280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34494848))), shape = tensor<uint32, [4]>([768, 3072, 1, 1])]; |
| 828 | tensor<fp16, [1, 768, 1, 1500]> var_1270_cast_fp16 = conv(dilations = var_1270_dilations_0, groups = var_1270_groups_0, pad = var_1270_pad_0, pad_type = var_1270_pad_type_0, strides = var_1270_strides_0, weight = layers_5_fc2_outlier_module_weight_to_fp16_sparsified, x = input_47_cast_fp16)[name = tensor<string, []>("op_1270_cast_fp16")]; |
| 829 | tensor<fp16, [1, 768, 1, 1500]> hidden_states_15_cast_fp16 = add(x = var_1264_cast_fp16, y = var_1270_cast_fp16)[name = tensor<string, []>("hidden_states_15_cast_fp16")]; |
| 830 | tensor<fp16, [1, 768, 1, 1500]> inputs_25_cast_fp16 = add(x = inputs_23_cast_fp16, y = hidden_states_15_cast_fp16)[name = tensor<string, []>("inputs_25_cast_fp16")]; |
| 831 | tensor<int32, []> var_1276 = const()[name = tensor<string, []>("op_1276"), val = tensor<int32, []>(3)]; |
| 832 | tensor<int32, [1]> out_25_axes_0 = const()[name = tensor<string, []>("out_25_axes_0"), val = tensor<int32, [1]>([1])]; |
| 833 | tensor<fp16, []> var_1298_to_fp16 = const()[name = tensor<string, []>("op_1298_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 834 | tensor<fp16, [1, 768, 1, 1500]> out_25_cast_fp16 = layer_norm(axes = out_25_axes_0, epsilon = var_1298_to_fp16, x = inputs_25_cast_fp16)[name = tensor<string, []>("out_25_cast_fp16")]; |
| 835 | tensor<fp16, [768]> obj_25_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_25_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34834496)))]; |
| 836 | tensor<fp16, [768]> obj_25_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_25_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34836096)))]; |
| 837 | tensor<fp16, []> obj_25_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_25_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 838 | tensor<fp16, [1, 768, 1, 1500]> obj_25_cast_fp16 = batch_norm(beta = obj_25_beta_0_to_fp16, epsilon = obj_25_epsilon_0_to_fp16, gamma = obj_25_gamma_0_to_fp16, mean = var_57_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_25_cast_fp16)[name = tensor<string, []>("obj_25_cast_fp16")]; |
| 839 | tensor<string, []> var_1320_pad_type_0 = const()[name = tensor<string, []>("op_1320_pad_type_0"), val = tensor<string, []>("valid")]; |
| 840 | tensor<int32, [2]> var_1320_strides_0 = const()[name = tensor<string, []>("op_1320_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 841 | tensor<int32, [4]> var_1320_pad_0 = const()[name = tensor<string, []>("op_1320_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 842 | tensor<int32, [2]> var_1320_dilations_0 = const()[name = tensor<string, []>("op_1320_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 843 | tensor<int32, []> var_1320_groups_0 = const()[name = tensor<string, []>("op_1320_groups_0"), val = tensor<int32, []>(1)]; |
| 844 | tensor<fp16, [768, 768, 1, 1]> layers_6_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34837696))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35132672))), name = tensor<string, []>("layers_6_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 845 | tensor<fp16, [768]> layers_6_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_6_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35132800)))]; |
| 846 | tensor<fp16, [1, 768, 1, 1500]> var_1320_cast_fp16 = conv(bias = layers_6_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_1320_dilations_0, groups = var_1320_groups_0, pad = var_1320_pad_0, pad_type = var_1320_pad_type_0, strides = var_1320_strides_0, weight = layers_6_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_25_cast_fp16)[name = tensor<string, []>("op_1320_cast_fp16")]; |
| 847 | tensor<string, []> var_1326_pad_type_0 = const()[name = tensor<string, []>("op_1326_pad_type_0"), val = tensor<string, []>("valid")]; |
| 848 | tensor<int32, [2]> var_1326_strides_0 = const()[name = tensor<string, []>("op_1326_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 849 | tensor<int32, [4]> var_1326_pad_0 = const()[name = tensor<string, []>("op_1326_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 850 | tensor<int32, [2]> var_1326_dilations_0 = const()[name = tensor<string, []>("op_1326_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 851 | tensor<int32, []> var_1326_groups_0 = const()[name = tensor<string, []>("op_1326_groups_0"), val = tensor<int32, []>(1)]; |
| 852 | tensor<fp16, [768, 768, 1, 1]> layers_6_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35142208))), name = tensor<string, []>("layers_6_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [3870]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35134400))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 853 | tensor<fp16, [1, 768, 1, 1500]> var_1326_cast_fp16 = conv(dilations = var_1326_dilations_0, groups = var_1326_groups_0, pad = var_1326_pad_0, pad_type = var_1326_pad_type_0, strides = var_1326_strides_0, weight = layers_6_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_25_cast_fp16)[name = tensor<string, []>("op_1326_cast_fp16")]; |
| 854 | tensor<fp16, [1, 768, 1, 1500]> query_13_cast_fp16 = add(x = var_1320_cast_fp16, y = var_1326_cast_fp16)[name = tensor<string, []>("query_13_cast_fp16")]; |
| 855 | tensor<string, []> var_1335_pad_type_0 = const()[name = tensor<string, []>("op_1335_pad_type_0"), val = tensor<string, []>("valid")]; |
| 856 | tensor<int32, [2]> var_1335_strides_0 = const()[name = tensor<string, []>("op_1335_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 857 | tensor<int32, [4]> var_1335_pad_0 = const()[name = tensor<string, []>("op_1335_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 858 | tensor<int32, [2]> var_1335_dilations_0 = const()[name = tensor<string, []>("op_1335_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 859 | tensor<int32, []> var_1335_groups_0 = const()[name = tensor<string, []>("op_1335_groups_0"), val = tensor<int32, []>(1)]; |
| 860 | tensor<fp16, [768, 768, 1, 1]> layers_6_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35216000))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35510976))), name = tensor<string, []>("layers_6_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 861 | tensor<fp16, [1, 768, 1, 1500]> var_1335_cast_fp16 = conv(dilations = var_1335_dilations_0, groups = var_1335_groups_0, pad = var_1335_pad_0, pad_type = var_1335_pad_type_0, strides = var_1335_strides_0, weight = layers_6_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_25_cast_fp16)[name = tensor<string, []>("op_1335_cast_fp16")]; |
| 862 | tensor<string, []> var_1341_pad_type_0 = const()[name = tensor<string, []>("op_1341_pad_type_0"), val = tensor<string, []>("valid")]; |
| 863 | tensor<int32, [2]> var_1341_strides_0 = const()[name = tensor<string, []>("op_1341_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 864 | tensor<int32, [4]> var_1341_pad_0 = const()[name = tensor<string, []>("op_1341_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 865 | tensor<int32, [2]> var_1341_dilations_0 = const()[name = tensor<string, []>("op_1341_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 866 | tensor<int32, []> var_1341_groups_0 = const()[name = tensor<string, []>("op_1341_groups_0"), val = tensor<int32, []>(1)]; |
| 867 | tensor<fp16, [768, 768, 1, 1]> layers_6_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35518848))), name = tensor<string, []>("layers_6_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [3809]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35511104))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 868 | tensor<fp16, [1, 768, 1, 1500]> var_1341_cast_fp16 = conv(dilations = var_1341_dilations_0, groups = var_1341_groups_0, pad = var_1341_pad_0, pad_type = var_1341_pad_type_0, strides = var_1341_strides_0, weight = layers_6_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_25_cast_fp16)[name = tensor<string, []>("op_1341_cast_fp16")]; |
| 869 | tensor<fp16, [1, 768, 1, 1500]> key_13_cast_fp16 = add(x = var_1335_cast_fp16, y = var_1341_cast_fp16)[name = tensor<string, []>("key_13_cast_fp16")]; |
| 870 | tensor<string, []> var_1351_pad_type_0 = const()[name = tensor<string, []>("op_1351_pad_type_0"), val = tensor<string, []>("valid")]; |
| 871 | tensor<int32, [2]> var_1351_strides_0 = const()[name = tensor<string, []>("op_1351_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 872 | tensor<int32, [4]> var_1351_pad_0 = const()[name = tensor<string, []>("op_1351_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 873 | tensor<int32, [2]> var_1351_dilations_0 = const()[name = tensor<string, []>("op_1351_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 874 | tensor<int32, []> var_1351_groups_0 = const()[name = tensor<string, []>("op_1351_groups_0"), val = tensor<int32, []>(1)]; |
| 875 | tensor<fp16, [768, 768, 1, 1]> layers_6_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35592640))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35887616))), name = tensor<string, []>("layers_6_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 876 | tensor<fp16, [768]> layers_6_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_6_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35887744)))]; |
| 877 | tensor<fp16, [1, 768, 1, 1500]> var_1351_cast_fp16 = conv(bias = layers_6_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_1351_dilations_0, groups = var_1351_groups_0, pad = var_1351_pad_0, pad_type = var_1351_pad_type_0, strides = var_1351_strides_0, weight = layers_6_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_25_cast_fp16)[name = tensor<string, []>("op_1351_cast_fp16")]; |
| 878 | tensor<string, []> var_1357_pad_type_0 = const()[name = tensor<string, []>("op_1357_pad_type_0"), val = tensor<string, []>("valid")]; |
| 879 | tensor<int32, [2]> var_1357_strides_0 = const()[name = tensor<string, []>("op_1357_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 880 | tensor<int32, [4]> var_1357_pad_0 = const()[name = tensor<string, []>("op_1357_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 881 | tensor<int32, [2]> var_1357_dilations_0 = const()[name = tensor<string, []>("op_1357_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 882 | tensor<int32, []> var_1357_groups_0 = const()[name = tensor<string, []>("op_1357_groups_0"), val = tensor<int32, []>(1)]; |
| 883 | tensor<fp16, [768, 768, 1, 1]> layers_6_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35895296))), name = tensor<string, []>("layers_6_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [2927]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35889344))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 884 | tensor<fp16, [1, 768, 1, 1500]> var_1357_cast_fp16 = conv(dilations = var_1357_dilations_0, groups = var_1357_groups_0, pad = var_1357_pad_0, pad_type = var_1357_pad_type_0, strides = var_1357_strides_0, weight = layers_6_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_25_cast_fp16)[name = tensor<string, []>("op_1357_cast_fp16")]; |
| 885 | tensor<fp16, [1, 768, 1, 1500]> value_13_cast_fp16 = add(x = var_1351_cast_fp16, y = var_1357_cast_fp16)[name = tensor<string, []>("value_13_cast_fp16")]; |
| 886 | tensor<int32, [4]> var_1361 = const()[name = tensor<string, []>("op_1361"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 887 | tensor<fp16, [1, 12, 64, 1500]> mh_q_13_cast_fp16 = reshape(shape = var_1361, x = query_13_cast_fp16)[name = tensor<string, []>("mh_q_13_cast_fp16")]; |
| 888 | tensor<fp16, []> var_1363_to_fp16 = const()[name = tensor<string, []>("op_1363_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| 889 | tensor<fp16, [1, 12, 64, 1500]> var_1364_cast_fp16 = mul(x = mh_q_13_cast_fp16, y = var_1363_to_fp16)[name = tensor<string, []>("op_1364_cast_fp16")]; |
| 890 | tensor<int32, [4]> var_1367 = const()[name = tensor<string, []>("op_1367"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 891 | tensor<fp16, [1, 12, 64, 1500]> var_1368_cast_fp16 = reshape(shape = var_1367, x = key_13_cast_fp16)[name = tensor<string, []>("op_1368_cast_fp16")]; |
| 892 | tensor<bool, []> mh_w_13_transpose_x_0 = const()[name = tensor<string, []>("mh_w_13_transpose_x_0"), val = tensor<bool, []>(true)]; |
| 893 | tensor<bool, []> mh_w_13_transpose_y_0 = const()[name = tensor<string, []>("mh_w_13_transpose_y_0"), val = tensor<bool, []>(false)]; |
| 894 | tensor<fp16, [1, 12, 1500, 1500]> mh_w_13_cast_fp16 = matmul(transpose_x = mh_w_13_transpose_x_0, transpose_y = mh_w_13_transpose_y_0, x = var_1364_cast_fp16, y = var_1368_cast_fp16)[name = tensor<string, []>("mh_w_13_cast_fp16")]; |
| 895 | tensor<fp16, [1, 12, 1500, 1500]> var_1371_cast_fp16 = softmax(axis = var_1276, x = mh_w_13_cast_fp16)[name = tensor<string, []>("op_1371_cast_fp16")]; |
| 896 | tensor<int32, [4]> var_1372 = const()[name = tensor<string, []>("op_1372"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 897 | tensor<fp16, [1, 12, 64, 1500]> var_1373_cast_fp16 = reshape(shape = var_1372, x = value_13_cast_fp16)[name = tensor<string, []>("op_1373_cast_fp16")]; |
| 898 | tensor<bool, []> attn_13_transpose_x_0 = const()[name = tensor<string, []>("attn_13_transpose_x_0"), val = tensor<bool, []>(false)]; |
| 899 | tensor<bool, []> attn_13_transpose_y_0 = const()[name = tensor<string, []>("attn_13_transpose_y_0"), val = tensor<bool, []>(true)]; |
| 900 | tensor<fp16, [1, 12, 64, 1500]> attn_13_cast_fp16 = matmul(transpose_x = attn_13_transpose_x_0, transpose_y = attn_13_transpose_y_0, x = var_1373_cast_fp16, y = var_1371_cast_fp16)[name = tensor<string, []>("attn_13_cast_fp16")]; |
| 901 | tensor<int32, [4]> var_1376 = const()[name = tensor<string, []>("op_1376"), val = tensor<int32, [4]>([1, 768, 1, 1500])]; |
| 902 | tensor<fp16, [1, 768, 1, 1500]> input_49_cast_fp16 = reshape(shape = var_1376, x = attn_13_cast_fp16)[name = tensor<string, []>("input_49_cast_fp16")]; |
| 903 | tensor<string, []> var_1386_pad_type_0 = const()[name = tensor<string, []>("op_1386_pad_type_0"), val = tensor<string, []>("valid")]; |
| 904 | tensor<int32, [2]> var_1386_strides_0 = const()[name = tensor<string, []>("op_1386_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 905 | tensor<int32, [4]> var_1386_pad_0 = const()[name = tensor<string, []>("op_1386_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 906 | tensor<int32, [2]> var_1386_dilations_0 = const()[name = tensor<string, []>("op_1386_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 907 | tensor<int32, []> var_1386_groups_0 = const()[name = tensor<string, []>("op_1386_groups_0"), val = tensor<int32, []>(1)]; |
| 908 | tensor<fp16, [768, 768, 1, 1]> layers_6_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35969088))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36264064))), name = tensor<string, []>("layers_6_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 909 | tensor<fp16, [768]> layers_6_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_6_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36264192)))]; |
| 910 | tensor<fp16, [1, 768, 1, 1500]> var_1386_cast_fp16 = conv(bias = layers_6_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_1386_dilations_0, groups = var_1386_groups_0, pad = var_1386_pad_0, pad_type = var_1386_pad_type_0, strides = var_1386_strides_0, weight = layers_6_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_49_cast_fp16)[name = tensor<string, []>("op_1386_cast_fp16")]; |
| 911 | tensor<string, []> var_1392_pad_type_0 = const()[name = tensor<string, []>("op_1392_pad_type_0"), val = tensor<string, []>("valid")]; |
| 912 | tensor<int32, [2]> var_1392_strides_0 = const()[name = tensor<string, []>("op_1392_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 913 | tensor<int32, [4]> var_1392_pad_0 = const()[name = tensor<string, []>("op_1392_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 914 | tensor<int32, [2]> var_1392_dilations_0 = const()[name = tensor<string, []>("op_1392_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 915 | tensor<int32, []> var_1392_groups_0 = const()[name = tensor<string, []>("op_1392_groups_0"), val = tensor<int32, []>(1)]; |
| 916 | tensor<fp16, [768, 768, 1, 1]> layers_6_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36270720))), name = tensor<string, []>("layers_6_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [2412]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36265792))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 917 | tensor<fp16, [1, 768, 1, 1500]> var_1392_cast_fp16 = conv(dilations = var_1392_dilations_0, groups = var_1392_groups_0, pad = var_1392_pad_0, pad_type = var_1392_pad_type_0, strides = var_1392_strides_0, weight = layers_6_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_49_cast_fp16)[name = tensor<string, []>("op_1392_cast_fp16")]; |
| 918 | tensor<fp16, [1, 768, 1, 1500]> obj_27_cast_fp16 = add(x = var_1386_cast_fp16, y = var_1392_cast_fp16)[name = tensor<string, []>("obj_27_cast_fp16")]; |
| 919 | tensor<fp16, [1, 768, 1, 1500]> inputs_27_cast_fp16 = add(x = inputs_25_cast_fp16, y = obj_27_cast_fp16)[name = tensor<string, []>("inputs_27_cast_fp16")]; |
| 920 | tensor<int32, [1]> out_27_axes_0 = const()[name = tensor<string, []>("out_27_axes_0"), val = tensor<int32, [1]>([1])]; |
| 921 | tensor<fp16, []> var_1403_to_fp16 = const()[name = tensor<string, []>("op_1403_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 922 | tensor<fp16, [1, 768, 1, 1500]> out_27_cast_fp16 = layer_norm(axes = out_27_axes_0, epsilon = var_1403_to_fp16, x = inputs_27_cast_fp16)[name = tensor<string, []>("out_27_cast_fp16")]; |
| 923 | tensor<fp16, [768]> input_51_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_51_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36344512)))]; |
| 924 | tensor<fp16, [768]> input_51_beta_0_to_fp16 = const()[name = tensor<string, []>("input_51_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36346112)))]; |
| 925 | tensor<fp16, []> input_51_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_51_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 926 | tensor<fp16, [1, 768, 1, 1500]> input_51_cast_fp16 = batch_norm(beta = input_51_beta_0_to_fp16, epsilon = input_51_epsilon_0_to_fp16, gamma = input_51_gamma_0_to_fp16, mean = var_57_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_27_cast_fp16)[name = tensor<string, []>("input_51_cast_fp16")]; |
| 927 | tensor<string, []> var_1421_pad_type_0 = const()[name = tensor<string, []>("op_1421_pad_type_0"), val = tensor<string, []>("valid")]; |
| 928 | tensor<int32, [2]> var_1421_strides_0 = const()[name = tensor<string, []>("op_1421_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 929 | tensor<int32, [4]> var_1421_pad_0 = const()[name = tensor<string, []>("op_1421_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 930 | tensor<int32, [2]> var_1421_dilations_0 = const()[name = tensor<string, []>("op_1421_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 931 | tensor<int32, []> var_1421_groups_0 = const()[name = tensor<string, []>("op_1421_groups_0"), val = tensor<int32, []>(1)]; |
| 932 | tensor<fp16, [3072, 768, 1, 1]> layers_6_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1179648]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36347712))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37527424))), name = tensor<string, []>("layers_6_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([3072, 768, 1, 1])]; |
| 933 | tensor<fp16, [3072]> layers_6_fc1_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_6_fc1_inlier_module_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37527552)))]; |
| 934 | tensor<fp16, [1, 3072, 1, 1500]> var_1421_cast_fp16 = conv(bias = layers_6_fc1_inlier_module_bias_to_fp16, dilations = var_1421_dilations_0, groups = var_1421_groups_0, pad = var_1421_pad_0, pad_type = var_1421_pad_type_0, strides = var_1421_strides_0, weight = layers_6_fc1_inlier_module_weight_to_fp16_palettized, x = input_51_cast_fp16)[name = tensor<string, []>("op_1421_cast_fp16")]; |
| 935 | tensor<string, []> var_1427_pad_type_0 = const()[name = tensor<string, []>("op_1427_pad_type_0"), val = tensor<string, []>("valid")]; |
| 936 | tensor<int32, [2]> var_1427_strides_0 = const()[name = tensor<string, []>("op_1427_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 937 | tensor<int32, [4]> var_1427_pad_0 = const()[name = tensor<string, []>("op_1427_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 938 | tensor<int32, [2]> var_1427_dilations_0 = const()[name = tensor<string, []>("op_1427_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 939 | tensor<int32, []> var_1427_groups_0 = const()[name = tensor<string, []>("op_1427_groups_0"), val = tensor<int32, []>(1)]; |
| 940 | tensor<fp16, [3072, 768, 1, 1]> layers_6_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37566848))), name = tensor<string, []>("layers_6_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [16482]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37533760))), shape = tensor<uint32, [4]>([3072, 768, 1, 1])]; |
| 941 | tensor<fp16, [1, 3072, 1, 1500]> var_1427_cast_fp16 = conv(dilations = var_1427_dilations_0, groups = var_1427_groups_0, pad = var_1427_pad_0, pad_type = var_1427_pad_type_0, strides = var_1427_strides_0, weight = layers_6_fc1_outlier_module_weight_to_fp16_sparsified, x = input_51_cast_fp16)[name = tensor<string, []>("op_1427_cast_fp16")]; |
| 942 | tensor<fp16, [1, 3072, 1, 1500]> input_53_cast_fp16 = add(x = var_1421_cast_fp16, y = var_1427_cast_fp16)[name = tensor<string, []>("input_53_cast_fp16")]; |
| 943 | tensor<string, []> input_55_mode_0 = const()[name = tensor<string, []>("input_55_mode_0"), val = tensor<string, []>("EXACT")]; |
| 944 | tensor<fp16, [1, 3072, 1, 1500]> input_55_cast_fp16 = gelu(mode = input_55_mode_0, x = input_53_cast_fp16)[name = tensor<string, []>("input_55_cast_fp16")]; |
| 945 | tensor<string, []> var_1438_pad_type_0 = const()[name = tensor<string, []>("op_1438_pad_type_0"), val = tensor<string, []>("valid")]; |
| 946 | tensor<int32, [2]> var_1438_strides_0 = const()[name = tensor<string, []>("op_1438_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 947 | tensor<int32, [4]> var_1438_pad_0 = const()[name = tensor<string, []>("op_1438_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 948 | tensor<int32, [2]> var_1438_dilations_0 = const()[name = tensor<string, []>("op_1438_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 949 | tensor<int32, []> var_1438_groups_0 = const()[name = tensor<string, []>("op_1438_groups_0"), val = tensor<int32, []>(1)]; |
| 950 | tensor<fp16, [768, 3072, 1, 1]> layers_6_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1179648]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37861824))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39041536))), name = tensor<string, []>("layers_6_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 3072, 1, 1])]; |
| 951 | tensor<fp16, [768]> layers_6_fc2_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_6_fc2_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39041664)))]; |
| 952 | tensor<fp16, [1, 768, 1, 1500]> var_1438_cast_fp16 = conv(bias = layers_6_fc2_inlier_module_bias_to_fp16, dilations = var_1438_dilations_0, groups = var_1438_groups_0, pad = var_1438_pad_0, pad_type = var_1438_pad_type_0, strides = var_1438_strides_0, weight = layers_6_fc2_inlier_module_weight_to_fp16_palettized, x = input_55_cast_fp16)[name = tensor<string, []>("op_1438_cast_fp16")]; |
| 953 | tensor<string, []> var_1444_pad_type_0 = const()[name = tensor<string, []>("op_1444_pad_type_0"), val = tensor<string, []>("valid")]; |
| 954 | tensor<int32, [2]> var_1444_strides_0 = const()[name = tensor<string, []>("op_1444_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 955 | tensor<int32, [4]> var_1444_pad_0 = const()[name = tensor<string, []>("op_1444_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 956 | tensor<int32, [2]> var_1444_dilations_0 = const()[name = tensor<string, []>("op_1444_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 957 | tensor<int32, []> var_1444_groups_0 = const()[name = tensor<string, []>("op_1444_groups_0"), val = tensor<int32, []>(1)]; |
| 958 | tensor<fp16, [768, 3072, 1, 1]> layers_6_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39079168))), name = tensor<string, []>("layers_6_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [17897]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39043264))), shape = tensor<uint32, [4]>([768, 3072, 1, 1])]; |
| 959 | tensor<fp16, [1, 768, 1, 1500]> var_1444_cast_fp16 = conv(dilations = var_1444_dilations_0, groups = var_1444_groups_0, pad = var_1444_pad_0, pad_type = var_1444_pad_type_0, strides = var_1444_strides_0, weight = layers_6_fc2_outlier_module_weight_to_fp16_sparsified, x = input_55_cast_fp16)[name = tensor<string, []>("op_1444_cast_fp16")]; |
| 960 | tensor<fp16, [1, 768, 1, 1500]> hidden_states_17_cast_fp16 = add(x = var_1438_cast_fp16, y = var_1444_cast_fp16)[name = tensor<string, []>("hidden_states_17_cast_fp16")]; |
| 961 | tensor<fp16, [1, 768, 1, 1500]> inputs_29_cast_fp16 = add(x = inputs_27_cast_fp16, y = hidden_states_17_cast_fp16)[name = tensor<string, []>("inputs_29_cast_fp16")]; |
| 962 | tensor<int32, []> var_1450 = const()[name = tensor<string, []>("op_1450"), val = tensor<int32, []>(3)]; |
| 963 | tensor<int32, [1]> out_29_axes_0 = const()[name = tensor<string, []>("out_29_axes_0"), val = tensor<int32, [1]>([1])]; |
| 964 | tensor<fp16, []> var_1472_to_fp16 = const()[name = tensor<string, []>("op_1472_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 965 | tensor<fp16, [1, 768, 1, 1500]> out_29_cast_fp16 = layer_norm(axes = out_29_axes_0, epsilon = var_1472_to_fp16, x = inputs_29_cast_fp16)[name = tensor<string, []>("out_29_cast_fp16")]; |
| 966 | tensor<fp16, [768]> obj_29_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_29_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39374144)))]; |
| 967 | tensor<fp16, [768]> obj_29_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_29_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39375744)))]; |
| 968 | tensor<fp16, []> obj_29_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_29_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 969 | tensor<fp16, [1, 768, 1, 1500]> obj_29_cast_fp16 = batch_norm(beta = obj_29_beta_0_to_fp16, epsilon = obj_29_epsilon_0_to_fp16, gamma = obj_29_gamma_0_to_fp16, mean = var_57_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_29_cast_fp16)[name = tensor<string, []>("obj_29_cast_fp16")]; |
| 970 | tensor<string, []> var_1494_pad_type_0 = const()[name = tensor<string, []>("op_1494_pad_type_0"), val = tensor<string, []>("valid")]; |
| 971 | tensor<int32, [2]> var_1494_strides_0 = const()[name = tensor<string, []>("op_1494_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 972 | tensor<int32, [4]> var_1494_pad_0 = const()[name = tensor<string, []>("op_1494_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 973 | tensor<int32, [2]> var_1494_dilations_0 = const()[name = tensor<string, []>("op_1494_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 974 | tensor<int32, []> var_1494_groups_0 = const()[name = tensor<string, []>("op_1494_groups_0"), val = tensor<int32, []>(1)]; |
| 975 | tensor<fp16, [768, 768, 1, 1]> layers_7_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39377344))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39672320))), name = tensor<string, []>("layers_7_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 976 | tensor<fp16, [768]> layers_7_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_7_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39672448)))]; |
| 977 | tensor<fp16, [1, 768, 1, 1500]> var_1494_cast_fp16 = conv(bias = layers_7_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_1494_dilations_0, groups = var_1494_groups_0, pad = var_1494_pad_0, pad_type = var_1494_pad_type_0, strides = var_1494_strides_0, weight = layers_7_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_29_cast_fp16)[name = tensor<string, []>("op_1494_cast_fp16")]; |
| 978 | tensor<string, []> var_1500_pad_type_0 = const()[name = tensor<string, []>("op_1500_pad_type_0"), val = tensor<string, []>("valid")]; |
| 979 | tensor<int32, [2]> var_1500_strides_0 = const()[name = tensor<string, []>("op_1500_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 980 | tensor<int32, [4]> var_1500_pad_0 = const()[name = tensor<string, []>("op_1500_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 981 | tensor<int32, [2]> var_1500_dilations_0 = const()[name = tensor<string, []>("op_1500_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 982 | tensor<int32, []> var_1500_groups_0 = const()[name = tensor<string, []>("op_1500_groups_0"), val = tensor<int32, []>(1)]; |
| 983 | tensor<fp16, [768, 768, 1, 1]> layers_7_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39680704))), name = tensor<string, []>("layers_7_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [3266]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39674048))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 984 | tensor<fp16, [1, 768, 1, 1500]> var_1500_cast_fp16 = conv(dilations = var_1500_dilations_0, groups = var_1500_groups_0, pad = var_1500_pad_0, pad_type = var_1500_pad_type_0, strides = var_1500_strides_0, weight = layers_7_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_29_cast_fp16)[name = tensor<string, []>("op_1500_cast_fp16")]; |
| 985 | tensor<fp16, [1, 768, 1, 1500]> query_15_cast_fp16 = add(x = var_1494_cast_fp16, y = var_1500_cast_fp16)[name = tensor<string, []>("query_15_cast_fp16")]; |
| 986 | tensor<string, []> var_1509_pad_type_0 = const()[name = tensor<string, []>("op_1509_pad_type_0"), val = tensor<string, []>("valid")]; |
| 987 | tensor<int32, [2]> var_1509_strides_0 = const()[name = tensor<string, []>("op_1509_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 988 | tensor<int32, [4]> var_1509_pad_0 = const()[name = tensor<string, []>("op_1509_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 989 | tensor<int32, [2]> var_1509_dilations_0 = const()[name = tensor<string, []>("op_1509_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 990 | tensor<int32, []> var_1509_groups_0 = const()[name = tensor<string, []>("op_1509_groups_0"), val = tensor<int32, []>(1)]; |
| 991 | tensor<fp16, [768, 768, 1, 1]> layers_7_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39754496))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40049472))), name = tensor<string, []>("layers_7_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 992 | tensor<fp16, [1, 768, 1, 1500]> var_1509_cast_fp16 = conv(dilations = var_1509_dilations_0, groups = var_1509_groups_0, pad = var_1509_pad_0, pad_type = var_1509_pad_type_0, strides = var_1509_strides_0, weight = layers_7_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_29_cast_fp16)[name = tensor<string, []>("op_1509_cast_fp16")]; |
| 993 | tensor<string, []> var_1515_pad_type_0 = const()[name = tensor<string, []>("op_1515_pad_type_0"), val = tensor<string, []>("valid")]; |
| 994 | tensor<int32, [2]> var_1515_strides_0 = const()[name = tensor<string, []>("op_1515_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 995 | tensor<int32, [4]> var_1515_pad_0 = const()[name = tensor<string, []>("op_1515_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 996 | tensor<int32, [2]> var_1515_dilations_0 = const()[name = tensor<string, []>("op_1515_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 997 | tensor<int32, []> var_1515_groups_0 = const()[name = tensor<string, []>("op_1515_groups_0"), val = tensor<int32, []>(1)]; |
| 998 | tensor<fp16, [768, 768, 1, 1]> layers_7_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40056832))), name = tensor<string, []>("layers_7_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [3558]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40049600))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 999 | tensor<fp16, [1, 768, 1, 1500]> var_1515_cast_fp16 = conv(dilations = var_1515_dilations_0, groups = var_1515_groups_0, pad = var_1515_pad_0, pad_type = var_1515_pad_type_0, strides = var_1515_strides_0, weight = layers_7_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_29_cast_fp16)[name = tensor<string, []>("op_1515_cast_fp16")]; |
| 1000 | tensor<fp16, [1, 768, 1, 1500]> key_15_cast_fp16 = add(x = var_1509_cast_fp16, y = var_1515_cast_fp16)[name = tensor<string, []>("key_15_cast_fp16")]; |
| 1001 | tensor<string, []> var_1525_pad_type_0 = const()[name = tensor<string, []>("op_1525_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1002 | tensor<int32, [2]> var_1525_strides_0 = const()[name = tensor<string, []>("op_1525_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1003 | tensor<int32, [4]> var_1525_pad_0 = const()[name = tensor<string, []>("op_1525_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1004 | tensor<int32, [2]> var_1525_dilations_0 = const()[name = tensor<string, []>("op_1525_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1005 | tensor<int32, []> var_1525_groups_0 = const()[name = tensor<string, []>("op_1525_groups_0"), val = tensor<int32, []>(1)]; |
| 1006 | tensor<fp16, [768, 768, 1, 1]> layers_7_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40130624))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40425600))), name = tensor<string, []>("layers_7_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1007 | tensor<fp16, [768]> layers_7_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_7_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40425728)))]; |
| 1008 | tensor<fp16, [1, 768, 1, 1500]> var_1525_cast_fp16 = conv(bias = layers_7_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_1525_dilations_0, groups = var_1525_groups_0, pad = var_1525_pad_0, pad_type = var_1525_pad_type_0, strides = var_1525_strides_0, weight = layers_7_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_29_cast_fp16)[name = tensor<string, []>("op_1525_cast_fp16")]; |
| 1009 | tensor<string, []> var_1531_pad_type_0 = const()[name = tensor<string, []>("op_1531_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1010 | tensor<int32, [2]> var_1531_strides_0 = const()[name = tensor<string, []>("op_1531_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1011 | tensor<int32, [4]> var_1531_pad_0 = const()[name = tensor<string, []>("op_1531_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1012 | tensor<int32, [2]> var_1531_dilations_0 = const()[name = tensor<string, []>("op_1531_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1013 | tensor<int32, []> var_1531_groups_0 = const()[name = tensor<string, []>("op_1531_groups_0"), val = tensor<int32, []>(1)]; |
| 1014 | tensor<fp16, [768, 768, 1, 1]> layers_7_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40433728))), name = tensor<string, []>("layers_7_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [3154]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40427328))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1015 | tensor<fp16, [1, 768, 1, 1500]> var_1531_cast_fp16 = conv(dilations = var_1531_dilations_0, groups = var_1531_groups_0, pad = var_1531_pad_0, pad_type = var_1531_pad_type_0, strides = var_1531_strides_0, weight = layers_7_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_29_cast_fp16)[name = tensor<string, []>("op_1531_cast_fp16")]; |
| 1016 | tensor<fp16, [1, 768, 1, 1500]> value_15_cast_fp16 = add(x = var_1525_cast_fp16, y = var_1531_cast_fp16)[name = tensor<string, []>("value_15_cast_fp16")]; |
| 1017 | tensor<int32, [4]> var_1535 = const()[name = tensor<string, []>("op_1535"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 1018 | tensor<fp16, [1, 12, 64, 1500]> mh_q_15_cast_fp16 = reshape(shape = var_1535, x = query_15_cast_fp16)[name = tensor<string, []>("mh_q_15_cast_fp16")]; |
| 1019 | tensor<fp16, []> var_1537_to_fp16 = const()[name = tensor<string, []>("op_1537_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| 1020 | tensor<fp16, [1, 12, 64, 1500]> var_1538_cast_fp16 = mul(x = mh_q_15_cast_fp16, y = var_1537_to_fp16)[name = tensor<string, []>("op_1538_cast_fp16")]; |
| 1021 | tensor<int32, [4]> var_1541 = const()[name = tensor<string, []>("op_1541"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 1022 | tensor<fp16, [1, 12, 64, 1500]> var_1542_cast_fp16 = reshape(shape = var_1541, x = key_15_cast_fp16)[name = tensor<string, []>("op_1542_cast_fp16")]; |
| 1023 | tensor<bool, []> mh_w_15_transpose_x_0 = const()[name = tensor<string, []>("mh_w_15_transpose_x_0"), val = tensor<bool, []>(true)]; |
| 1024 | tensor<bool, []> mh_w_15_transpose_y_0 = const()[name = tensor<string, []>("mh_w_15_transpose_y_0"), val = tensor<bool, []>(false)]; |
| 1025 | tensor<fp16, [1, 12, 1500, 1500]> mh_w_15_cast_fp16 = matmul(transpose_x = mh_w_15_transpose_x_0, transpose_y = mh_w_15_transpose_y_0, x = var_1538_cast_fp16, y = var_1542_cast_fp16)[name = tensor<string, []>("mh_w_15_cast_fp16")]; |
| 1026 | tensor<fp16, [1, 12, 1500, 1500]> var_1545_cast_fp16 = softmax(axis = var_1450, x = mh_w_15_cast_fp16)[name = tensor<string, []>("op_1545_cast_fp16")]; |
| 1027 | tensor<int32, [4]> var_1546 = const()[name = tensor<string, []>("op_1546"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 1028 | tensor<fp16, [1, 12, 64, 1500]> var_1547_cast_fp16 = reshape(shape = var_1546, x = value_15_cast_fp16)[name = tensor<string, []>("op_1547_cast_fp16")]; |
| 1029 | tensor<bool, []> attn_15_transpose_x_0 = const()[name = tensor<string, []>("attn_15_transpose_x_0"), val = tensor<bool, []>(false)]; |
| 1030 | tensor<bool, []> attn_15_transpose_y_0 = const()[name = tensor<string, []>("attn_15_transpose_y_0"), val = tensor<bool, []>(true)]; |
| 1031 | tensor<fp16, [1, 12, 64, 1500]> attn_15_cast_fp16 = matmul(transpose_x = attn_15_transpose_x_0, transpose_y = attn_15_transpose_y_0, x = var_1547_cast_fp16, y = var_1545_cast_fp16)[name = tensor<string, []>("attn_15_cast_fp16")]; |
| 1032 | tensor<int32, [4]> var_1550 = const()[name = tensor<string, []>("op_1550"), val = tensor<int32, [4]>([1, 768, 1, 1500])]; |
| 1033 | tensor<fp16, [1, 768, 1, 1500]> input_57_cast_fp16 = reshape(shape = var_1550, x = attn_15_cast_fp16)[name = tensor<string, []>("input_57_cast_fp16")]; |
| 1034 | tensor<string, []> var_1560_pad_type_0 = const()[name = tensor<string, []>("op_1560_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1035 | tensor<int32, [2]> var_1560_strides_0 = const()[name = tensor<string, []>("op_1560_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1036 | tensor<int32, [4]> var_1560_pad_0 = const()[name = tensor<string, []>("op_1560_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1037 | tensor<int32, [2]> var_1560_dilations_0 = const()[name = tensor<string, []>("op_1560_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1038 | tensor<int32, []> var_1560_groups_0 = const()[name = tensor<string, []>("op_1560_groups_0"), val = tensor<int32, []>(1)]; |
| 1039 | tensor<fp16, [768, 768, 1, 1]> layers_7_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40507520))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40802496))), name = tensor<string, []>("layers_7_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1040 | tensor<fp16, [768]> layers_7_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_7_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40802624)))]; |
| 1041 | tensor<fp16, [1, 768, 1, 1500]> var_1560_cast_fp16 = conv(bias = layers_7_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_1560_dilations_0, groups = var_1560_groups_0, pad = var_1560_pad_0, pad_type = var_1560_pad_type_0, strides = var_1560_strides_0, weight = layers_7_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_57_cast_fp16)[name = tensor<string, []>("op_1560_cast_fp16")]; |
| 1042 | tensor<string, []> var_1566_pad_type_0 = const()[name = tensor<string, []>("op_1566_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1043 | tensor<int32, [2]> var_1566_strides_0 = const()[name = tensor<string, []>("op_1566_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1044 | tensor<int32, [4]> var_1566_pad_0 = const()[name = tensor<string, []>("op_1566_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1045 | tensor<int32, [2]> var_1566_dilations_0 = const()[name = tensor<string, []>("op_1566_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1046 | tensor<int32, []> var_1566_groups_0 = const()[name = tensor<string, []>("op_1566_groups_0"), val = tensor<int32, []>(1)]; |
| 1047 | tensor<fp16, [768, 768, 1, 1]> layers_7_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40810112))), name = tensor<string, []>("layers_7_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [2902]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40804224))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1048 | tensor<fp16, [1, 768, 1, 1500]> var_1566_cast_fp16 = conv(dilations = var_1566_dilations_0, groups = var_1566_groups_0, pad = var_1566_pad_0, pad_type = var_1566_pad_type_0, strides = var_1566_strides_0, weight = layers_7_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_57_cast_fp16)[name = tensor<string, []>("op_1566_cast_fp16")]; |
| 1049 | tensor<fp16, [1, 768, 1, 1500]> obj_31_cast_fp16 = add(x = var_1560_cast_fp16, y = var_1566_cast_fp16)[name = tensor<string, []>("obj_31_cast_fp16")]; |
| 1050 | tensor<fp16, [1, 768, 1, 1500]> inputs_31_cast_fp16 = add(x = inputs_29_cast_fp16, y = obj_31_cast_fp16)[name = tensor<string, []>("inputs_31_cast_fp16")]; |
| 1051 | tensor<int32, [1]> out_31_axes_0 = const()[name = tensor<string, []>("out_31_axes_0"), val = tensor<int32, [1]>([1])]; |
| 1052 | tensor<fp16, []> var_1577_to_fp16 = const()[name = tensor<string, []>("op_1577_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 1053 | tensor<fp16, [1, 768, 1, 1500]> out_31_cast_fp16 = layer_norm(axes = out_31_axes_0, epsilon = var_1577_to_fp16, x = inputs_31_cast_fp16)[name = tensor<string, []>("out_31_cast_fp16")]; |
| 1054 | tensor<fp16, [768]> input_59_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_59_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40883904)))]; |
| 1055 | tensor<fp16, [768]> input_59_beta_0_to_fp16 = const()[name = tensor<string, []>("input_59_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40885504)))]; |
| 1056 | tensor<fp16, []> input_59_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_59_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 1057 | tensor<fp16, [1, 768, 1, 1500]> input_59_cast_fp16 = batch_norm(beta = input_59_beta_0_to_fp16, epsilon = input_59_epsilon_0_to_fp16, gamma = input_59_gamma_0_to_fp16, mean = var_57_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_31_cast_fp16)[name = tensor<string, []>("input_59_cast_fp16")]; |
| 1058 | tensor<string, []> var_1595_pad_type_0 = const()[name = tensor<string, []>("op_1595_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1059 | tensor<int32, [2]> var_1595_strides_0 = const()[name = tensor<string, []>("op_1595_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1060 | tensor<int32, [4]> var_1595_pad_0 = const()[name = tensor<string, []>("op_1595_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1061 | tensor<int32, [2]> var_1595_dilations_0 = const()[name = tensor<string, []>("op_1595_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1062 | tensor<int32, []> var_1595_groups_0 = const()[name = tensor<string, []>("op_1595_groups_0"), val = tensor<int32, []>(1)]; |
| 1063 | tensor<fp16, [3072, 768, 1, 1]> layers_7_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1179648]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40887104))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42066816))), name = tensor<string, []>("layers_7_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([3072, 768, 1, 1])]; |
| 1064 | tensor<fp16, [3072]> layers_7_fc1_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_7_fc1_inlier_module_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42066944)))]; |
| 1065 | tensor<fp16, [1, 3072, 1, 1500]> var_1595_cast_fp16 = conv(bias = layers_7_fc1_inlier_module_bias_to_fp16, dilations = var_1595_dilations_0, groups = var_1595_groups_0, pad = var_1595_pad_0, pad_type = var_1595_pad_type_0, strides = var_1595_strides_0, weight = layers_7_fc1_inlier_module_weight_to_fp16_palettized, x = input_59_cast_fp16)[name = tensor<string, []>("op_1595_cast_fp16")]; |
| 1066 | tensor<string, []> var_1601_pad_type_0 = const()[name = tensor<string, []>("op_1601_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1067 | tensor<int32, [2]> var_1601_strides_0 = const()[name = tensor<string, []>("op_1601_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1068 | tensor<int32, [4]> var_1601_pad_0 = const()[name = tensor<string, []>("op_1601_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1069 | tensor<int32, [2]> var_1601_dilations_0 = const()[name = tensor<string, []>("op_1601_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1070 | tensor<int32, []> var_1601_groups_0 = const()[name = tensor<string, []>("op_1601_groups_0"), val = tensor<int32, []>(1)]; |
| 1071 | tensor<fp16, [3072, 768, 1, 1]> layers_7_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42110976))), name = tensor<string, []>("layers_7_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [18849]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42073152))), shape = tensor<uint32, [4]>([3072, 768, 1, 1])]; |
| 1072 | tensor<fp16, [1, 3072, 1, 1500]> var_1601_cast_fp16 = conv(dilations = var_1601_dilations_0, groups = var_1601_groups_0, pad = var_1601_pad_0, pad_type = var_1601_pad_type_0, strides = var_1601_strides_0, weight = layers_7_fc1_outlier_module_weight_to_fp16_sparsified, x = input_59_cast_fp16)[name = tensor<string, []>("op_1601_cast_fp16")]; |
| 1073 | tensor<fp16, [1, 3072, 1, 1500]> input_61_cast_fp16 = add(x = var_1595_cast_fp16, y = var_1601_cast_fp16)[name = tensor<string, []>("input_61_cast_fp16")]; |
| 1074 | tensor<string, []> input_63_mode_0 = const()[name = tensor<string, []>("input_63_mode_0"), val = tensor<string, []>("EXACT")]; |
| 1075 | tensor<fp16, [1, 3072, 1, 1500]> input_63_cast_fp16 = gelu(mode = input_63_mode_0, x = input_61_cast_fp16)[name = tensor<string, []>("input_63_cast_fp16")]; |
| 1076 | tensor<string, []> var_1612_pad_type_0 = const()[name = tensor<string, []>("op_1612_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1077 | tensor<int32, [2]> var_1612_strides_0 = const()[name = tensor<string, []>("op_1612_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1078 | tensor<int32, [4]> var_1612_pad_0 = const()[name = tensor<string, []>("op_1612_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1079 | tensor<int32, [2]> var_1612_dilations_0 = const()[name = tensor<string, []>("op_1612_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1080 | tensor<int32, []> var_1612_groups_0 = const()[name = tensor<string, []>("op_1612_groups_0"), val = tensor<int32, []>(1)]; |
| 1081 | tensor<fp16, [768, 3072, 1, 1]> layers_7_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1179648]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42405952))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43585664))), name = tensor<string, []>("layers_7_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 3072, 1, 1])]; |
| 1082 | tensor<fp16, [768]> layers_7_fc2_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_7_fc2_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43585792)))]; |
| 1083 | tensor<fp16, [1, 768, 1, 1500]> var_1612_cast_fp16 = conv(bias = layers_7_fc2_inlier_module_bias_to_fp16, dilations = var_1612_dilations_0, groups = var_1612_groups_0, pad = var_1612_pad_0, pad_type = var_1612_pad_type_0, strides = var_1612_strides_0, weight = layers_7_fc2_inlier_module_weight_to_fp16_palettized, x = input_63_cast_fp16)[name = tensor<string, []>("op_1612_cast_fp16")]; |
| 1084 | tensor<string, []> var_1618_pad_type_0 = const()[name = tensor<string, []>("op_1618_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1085 | tensor<int32, [2]> var_1618_strides_0 = const()[name = tensor<string, []>("op_1618_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1086 | tensor<int32, [4]> var_1618_pad_0 = const()[name = tensor<string, []>("op_1618_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1087 | tensor<int32, [2]> var_1618_dilations_0 = const()[name = tensor<string, []>("op_1618_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1088 | tensor<int32, []> var_1618_groups_0 = const()[name = tensor<string, []>("op_1618_groups_0"), val = tensor<int32, []>(1)]; |
| 1089 | tensor<fp16, [768, 3072, 1, 1]> layers_7_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43616960))), name = tensor<string, []>("layers_7_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [14729]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43587392))), shape = tensor<uint32, [4]>([768, 3072, 1, 1])]; |
| 1090 | tensor<fp16, [1, 768, 1, 1500]> var_1618_cast_fp16 = conv(dilations = var_1618_dilations_0, groups = var_1618_groups_0, pad = var_1618_pad_0, pad_type = var_1618_pad_type_0, strides = var_1618_strides_0, weight = layers_7_fc2_outlier_module_weight_to_fp16_sparsified, x = input_63_cast_fp16)[name = tensor<string, []>("op_1618_cast_fp16")]; |
| 1091 | tensor<fp16, [1, 768, 1, 1500]> hidden_states_19_cast_fp16 = add(x = var_1612_cast_fp16, y = var_1618_cast_fp16)[name = tensor<string, []>("hidden_states_19_cast_fp16")]; |
| 1092 | tensor<fp16, [1, 768, 1, 1500]> inputs_33_cast_fp16 = add(x = inputs_31_cast_fp16, y = hidden_states_19_cast_fp16)[name = tensor<string, []>("inputs_33_cast_fp16")]; |
| 1093 | tensor<int32, []> var_1624 = const()[name = tensor<string, []>("op_1624"), val = tensor<int32, []>(3)]; |
| 1094 | tensor<int32, [1]> out_33_axes_0 = const()[name = tensor<string, []>("out_33_axes_0"), val = tensor<int32, [1]>([1])]; |
| 1095 | tensor<fp16, []> var_1646_to_fp16 = const()[name = tensor<string, []>("op_1646_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 1096 | tensor<fp16, [1, 768, 1, 1500]> out_33_cast_fp16 = layer_norm(axes = out_33_axes_0, epsilon = var_1646_to_fp16, x = inputs_33_cast_fp16)[name = tensor<string, []>("out_33_cast_fp16")]; |
| 1097 | tensor<fp16, [768]> obj_33_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_33_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43911936)))]; |
| 1098 | tensor<fp16, [768]> obj_33_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_33_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43913536)))]; |
| 1099 | tensor<fp16, []> obj_33_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_33_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 1100 | tensor<fp16, [1, 768, 1, 1500]> obj_33_cast_fp16 = batch_norm(beta = obj_33_beta_0_to_fp16, epsilon = obj_33_epsilon_0_to_fp16, gamma = obj_33_gamma_0_to_fp16, mean = var_57_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_33_cast_fp16)[name = tensor<string, []>("obj_33_cast_fp16")]; |
| 1101 | tensor<string, []> var_1668_pad_type_0 = const()[name = tensor<string, []>("op_1668_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1102 | tensor<int32, [2]> var_1668_strides_0 = const()[name = tensor<string, []>("op_1668_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1103 | tensor<int32, [4]> var_1668_pad_0 = const()[name = tensor<string, []>("op_1668_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1104 | tensor<int32, [2]> var_1668_dilations_0 = const()[name = tensor<string, []>("op_1668_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1105 | tensor<int32, []> var_1668_groups_0 = const()[name = tensor<string, []>("op_1668_groups_0"), val = tensor<int32, []>(1)]; |
| 1106 | tensor<fp16, [768, 768, 1, 1]> layers_8_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43915136))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44210112))), name = tensor<string, []>("layers_8_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1107 | tensor<fp16, [768]> layers_8_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_8_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44210240)))]; |
| 1108 | tensor<fp16, [1, 768, 1, 1500]> var_1668_cast_fp16 = conv(bias = layers_8_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_1668_dilations_0, groups = var_1668_groups_0, pad = var_1668_pad_0, pad_type = var_1668_pad_type_0, strides = var_1668_strides_0, weight = layers_8_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_33_cast_fp16)[name = tensor<string, []>("op_1668_cast_fp16")]; |
| 1109 | tensor<string, []> var_1674_pad_type_0 = const()[name = tensor<string, []>("op_1674_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1110 | tensor<int32, [2]> var_1674_strides_0 = const()[name = tensor<string, []>("op_1674_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1111 | tensor<int32, [4]> var_1674_pad_0 = const()[name = tensor<string, []>("op_1674_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1112 | tensor<int32, [2]> var_1674_dilations_0 = const()[name = tensor<string, []>("op_1674_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1113 | tensor<int32, []> var_1674_groups_0 = const()[name = tensor<string, []>("op_1674_groups_0"), val = tensor<int32, []>(1)]; |
| 1114 | tensor<fp16, [768, 768, 1, 1]> layers_8_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44219008))), name = tensor<string, []>("layers_8_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [3550]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44211840))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1115 | tensor<fp16, [1, 768, 1, 1500]> var_1674_cast_fp16 = conv(dilations = var_1674_dilations_0, groups = var_1674_groups_0, pad = var_1674_pad_0, pad_type = var_1674_pad_type_0, strides = var_1674_strides_0, weight = layers_8_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_33_cast_fp16)[name = tensor<string, []>("op_1674_cast_fp16")]; |
| 1116 | tensor<fp16, [1, 768, 1, 1500]> query_17_cast_fp16 = add(x = var_1668_cast_fp16, y = var_1674_cast_fp16)[name = tensor<string, []>("query_17_cast_fp16")]; |
| 1117 | tensor<string, []> var_1683_pad_type_0 = const()[name = tensor<string, []>("op_1683_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1118 | tensor<int32, [2]> var_1683_strides_0 = const()[name = tensor<string, []>("op_1683_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1119 | tensor<int32, [4]> var_1683_pad_0 = const()[name = tensor<string, []>("op_1683_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1120 | tensor<int32, [2]> var_1683_dilations_0 = const()[name = tensor<string, []>("op_1683_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1121 | tensor<int32, []> var_1683_groups_0 = const()[name = tensor<string, []>("op_1683_groups_0"), val = tensor<int32, []>(1)]; |
| 1122 | tensor<fp16, [768, 768, 1, 1]> layers_8_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44292800))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44587776))), name = tensor<string, []>("layers_8_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1123 | tensor<fp16, [1, 768, 1, 1500]> var_1683_cast_fp16 = conv(dilations = var_1683_dilations_0, groups = var_1683_groups_0, pad = var_1683_pad_0, pad_type = var_1683_pad_type_0, strides = var_1683_strides_0, weight = layers_8_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_33_cast_fp16)[name = tensor<string, []>("op_1683_cast_fp16")]; |
| 1124 | tensor<string, []> var_1689_pad_type_0 = const()[name = tensor<string, []>("op_1689_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1125 | tensor<int32, [2]> var_1689_strides_0 = const()[name = tensor<string, []>("op_1689_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1126 | tensor<int32, [4]> var_1689_pad_0 = const()[name = tensor<string, []>("op_1689_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1127 | tensor<int32, [2]> var_1689_dilations_0 = const()[name = tensor<string, []>("op_1689_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1128 | tensor<int32, []> var_1689_groups_0 = const()[name = tensor<string, []>("op_1689_groups_0"), val = tensor<int32, []>(1)]; |
| 1129 | tensor<fp16, [768, 768, 1, 1]> layers_8_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44595136))), name = tensor<string, []>("layers_8_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [3567]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44587904))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1130 | tensor<fp16, [1, 768, 1, 1500]> var_1689_cast_fp16 = conv(dilations = var_1689_dilations_0, groups = var_1689_groups_0, pad = var_1689_pad_0, pad_type = var_1689_pad_type_0, strides = var_1689_strides_0, weight = layers_8_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_33_cast_fp16)[name = tensor<string, []>("op_1689_cast_fp16")]; |
| 1131 | tensor<fp16, [1, 768, 1, 1500]> key_17_cast_fp16 = add(x = var_1683_cast_fp16, y = var_1689_cast_fp16)[name = tensor<string, []>("key_17_cast_fp16")]; |
| 1132 | tensor<string, []> var_1699_pad_type_0 = const()[name = tensor<string, []>("op_1699_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1133 | tensor<int32, [2]> var_1699_strides_0 = const()[name = tensor<string, []>("op_1699_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1134 | tensor<int32, [4]> var_1699_pad_0 = const()[name = tensor<string, []>("op_1699_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1135 | tensor<int32, [2]> var_1699_dilations_0 = const()[name = tensor<string, []>("op_1699_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1136 | tensor<int32, []> var_1699_groups_0 = const()[name = tensor<string, []>("op_1699_groups_0"), val = tensor<int32, []>(1)]; |
| 1137 | tensor<fp16, [768, 768, 1, 1]> layers_8_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44668928))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44963904))), name = tensor<string, []>("layers_8_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1138 | tensor<fp16, [768]> layers_8_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_8_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44964032)))]; |
| 1139 | tensor<fp16, [1, 768, 1, 1500]> var_1699_cast_fp16 = conv(bias = layers_8_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_1699_dilations_0, groups = var_1699_groups_0, pad = var_1699_pad_0, pad_type = var_1699_pad_type_0, strides = var_1699_strides_0, weight = layers_8_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_33_cast_fp16)[name = tensor<string, []>("op_1699_cast_fp16")]; |
| 1140 | tensor<string, []> var_1705_pad_type_0 = const()[name = tensor<string, []>("op_1705_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1141 | tensor<int32, [2]> var_1705_strides_0 = const()[name = tensor<string, []>("op_1705_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1142 | tensor<int32, [4]> var_1705_pad_0 = const()[name = tensor<string, []>("op_1705_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1143 | tensor<int32, [2]> var_1705_dilations_0 = const()[name = tensor<string, []>("op_1705_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1144 | tensor<int32, []> var_1705_groups_0 = const()[name = tensor<string, []>("op_1705_groups_0"), val = tensor<int32, []>(1)]; |
| 1145 | tensor<fp16, [768, 768, 1, 1]> layers_8_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44971200))), name = tensor<string, []>("layers_8_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [2744]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44965632))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1146 | tensor<fp16, [1, 768, 1, 1500]> var_1705_cast_fp16 = conv(dilations = var_1705_dilations_0, groups = var_1705_groups_0, pad = var_1705_pad_0, pad_type = var_1705_pad_type_0, strides = var_1705_strides_0, weight = layers_8_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_33_cast_fp16)[name = tensor<string, []>("op_1705_cast_fp16")]; |
| 1147 | tensor<fp16, [1, 768, 1, 1500]> value_17_cast_fp16 = add(x = var_1699_cast_fp16, y = var_1705_cast_fp16)[name = tensor<string, []>("value_17_cast_fp16")]; |
| 1148 | tensor<int32, [4]> var_1709 = const()[name = tensor<string, []>("op_1709"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 1149 | tensor<fp16, [1, 12, 64, 1500]> mh_q_17_cast_fp16 = reshape(shape = var_1709, x = query_17_cast_fp16)[name = tensor<string, []>("mh_q_17_cast_fp16")]; |
| 1150 | tensor<fp16, []> var_1711_to_fp16 = const()[name = tensor<string, []>("op_1711_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| 1151 | tensor<fp16, [1, 12, 64, 1500]> var_1712_cast_fp16 = mul(x = mh_q_17_cast_fp16, y = var_1711_to_fp16)[name = tensor<string, []>("op_1712_cast_fp16")]; |
| 1152 | tensor<int32, [4]> var_1715 = const()[name = tensor<string, []>("op_1715"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 1153 | tensor<fp16, [1, 12, 64, 1500]> var_1716_cast_fp16 = reshape(shape = var_1715, x = key_17_cast_fp16)[name = tensor<string, []>("op_1716_cast_fp16")]; |
| 1154 | tensor<bool, []> mh_w_17_transpose_x_0 = const()[name = tensor<string, []>("mh_w_17_transpose_x_0"), val = tensor<bool, []>(true)]; |
| 1155 | tensor<bool, []> mh_w_17_transpose_y_0 = const()[name = tensor<string, []>("mh_w_17_transpose_y_0"), val = tensor<bool, []>(false)]; |
| 1156 | tensor<fp16, [1, 12, 1500, 1500]> mh_w_17_cast_fp16 = matmul(transpose_x = mh_w_17_transpose_x_0, transpose_y = mh_w_17_transpose_y_0, x = var_1712_cast_fp16, y = var_1716_cast_fp16)[name = tensor<string, []>("mh_w_17_cast_fp16")]; |
| 1157 | tensor<fp16, [1, 12, 1500, 1500]> var_1719_cast_fp16 = softmax(axis = var_1624, x = mh_w_17_cast_fp16)[name = tensor<string, []>("op_1719_cast_fp16")]; |
| 1158 | tensor<int32, [4]> var_1720 = const()[name = tensor<string, []>("op_1720"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 1159 | tensor<fp16, [1, 12, 64, 1500]> var_1721_cast_fp16 = reshape(shape = var_1720, x = value_17_cast_fp16)[name = tensor<string, []>("op_1721_cast_fp16")]; |
| 1160 | tensor<bool, []> attn_17_transpose_x_0 = const()[name = tensor<string, []>("attn_17_transpose_x_0"), val = tensor<bool, []>(false)]; |
| 1161 | tensor<bool, []> attn_17_transpose_y_0 = const()[name = tensor<string, []>("attn_17_transpose_y_0"), val = tensor<bool, []>(true)]; |
| 1162 | tensor<fp16, [1, 12, 64, 1500]> attn_17_cast_fp16 = matmul(transpose_x = attn_17_transpose_x_0, transpose_y = attn_17_transpose_y_0, x = var_1721_cast_fp16, y = var_1719_cast_fp16)[name = tensor<string, []>("attn_17_cast_fp16")]; |
| 1163 | tensor<int32, [4]> var_1724 = const()[name = tensor<string, []>("op_1724"), val = tensor<int32, [4]>([1, 768, 1, 1500])]; |
| 1164 | tensor<fp16, [1, 768, 1, 1500]> input_65_cast_fp16 = reshape(shape = var_1724, x = attn_17_cast_fp16)[name = tensor<string, []>("input_65_cast_fp16")]; |
| 1165 | tensor<string, []> var_1734_pad_type_0 = const()[name = tensor<string, []>("op_1734_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1166 | tensor<int32, [2]> var_1734_strides_0 = const()[name = tensor<string, []>("op_1734_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1167 | tensor<int32, [4]> var_1734_pad_0 = const()[name = tensor<string, []>("op_1734_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1168 | tensor<int32, [2]> var_1734_dilations_0 = const()[name = tensor<string, []>("op_1734_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1169 | tensor<int32, []> var_1734_groups_0 = const()[name = tensor<string, []>("op_1734_groups_0"), val = tensor<int32, []>(1)]; |
| 1170 | tensor<fp16, [768, 768, 1, 1]> layers_8_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45044992))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45339968))), name = tensor<string, []>("layers_8_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1171 | tensor<fp16, [768]> layers_8_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_8_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45340096)))]; |
| 1172 | tensor<fp16, [1, 768, 1, 1500]> var_1734_cast_fp16 = conv(bias = layers_8_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_1734_dilations_0, groups = var_1734_groups_0, pad = var_1734_pad_0, pad_type = var_1734_pad_type_0, strides = var_1734_strides_0, weight = layers_8_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_65_cast_fp16)[name = tensor<string, []>("op_1734_cast_fp16")]; |
| 1173 | tensor<string, []> var_1740_pad_type_0 = const()[name = tensor<string, []>("op_1740_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1174 | tensor<int32, [2]> var_1740_strides_0 = const()[name = tensor<string, []>("op_1740_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1175 | tensor<int32, [4]> var_1740_pad_0 = const()[name = tensor<string, []>("op_1740_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1176 | tensor<int32, [2]> var_1740_dilations_0 = const()[name = tensor<string, []>("op_1740_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1177 | tensor<int32, []> var_1740_groups_0 = const()[name = tensor<string, []>("op_1740_groups_0"), val = tensor<int32, []>(1)]; |
| 1178 | tensor<fp16, [768, 768, 1, 1]> layers_8_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45347776))), name = tensor<string, []>("layers_8_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [2992]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45341696))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1179 | tensor<fp16, [1, 768, 1, 1500]> var_1740_cast_fp16 = conv(dilations = var_1740_dilations_0, groups = var_1740_groups_0, pad = var_1740_pad_0, pad_type = var_1740_pad_type_0, strides = var_1740_strides_0, weight = layers_8_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_65_cast_fp16)[name = tensor<string, []>("op_1740_cast_fp16")]; |
| 1180 | tensor<fp16, [1, 768, 1, 1500]> obj_35_cast_fp16 = add(x = var_1734_cast_fp16, y = var_1740_cast_fp16)[name = tensor<string, []>("obj_35_cast_fp16")]; |
| 1181 | tensor<fp16, [1, 768, 1, 1500]> inputs_35_cast_fp16 = add(x = inputs_33_cast_fp16, y = obj_35_cast_fp16)[name = tensor<string, []>("inputs_35_cast_fp16")]; |
| 1182 | tensor<int32, [1]> out_35_axes_0 = const()[name = tensor<string, []>("out_35_axes_0"), val = tensor<int32, [1]>([1])]; |
| 1183 | tensor<fp16, []> var_1751_to_fp16 = const()[name = tensor<string, []>("op_1751_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 1184 | tensor<fp16, [1, 768, 1, 1500]> out_35_cast_fp16 = layer_norm(axes = out_35_axes_0, epsilon = var_1751_to_fp16, x = inputs_35_cast_fp16)[name = tensor<string, []>("out_35_cast_fp16")]; |
| 1185 | tensor<fp16, [768]> input_67_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_67_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45421568)))]; |
| 1186 | tensor<fp16, [768]> input_67_beta_0_to_fp16 = const()[name = tensor<string, []>("input_67_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45423168)))]; |
| 1187 | tensor<fp16, []> input_67_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_67_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 1188 | tensor<fp16, [1, 768, 1, 1500]> input_67_cast_fp16 = batch_norm(beta = input_67_beta_0_to_fp16, epsilon = input_67_epsilon_0_to_fp16, gamma = input_67_gamma_0_to_fp16, mean = var_57_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_35_cast_fp16)[name = tensor<string, []>("input_67_cast_fp16")]; |
| 1189 | tensor<string, []> var_1769_pad_type_0 = const()[name = tensor<string, []>("op_1769_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1190 | tensor<int32, [2]> var_1769_strides_0 = const()[name = tensor<string, []>("op_1769_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1191 | tensor<int32, [4]> var_1769_pad_0 = const()[name = tensor<string, []>("op_1769_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1192 | tensor<int32, [2]> var_1769_dilations_0 = const()[name = tensor<string, []>("op_1769_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1193 | tensor<int32, []> var_1769_groups_0 = const()[name = tensor<string, []>("op_1769_groups_0"), val = tensor<int32, []>(1)]; |
| 1194 | tensor<fp16, [3072, 768, 1, 1]> layers_8_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1179648]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45424768))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46604480))), name = tensor<string, []>("layers_8_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([3072, 768, 1, 1])]; |
| 1195 | tensor<fp16, [3072]> layers_8_fc1_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_8_fc1_inlier_module_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46604608)))]; |
| 1196 | tensor<fp16, [1, 3072, 1, 1500]> var_1769_cast_fp16 = conv(bias = layers_8_fc1_inlier_module_bias_to_fp16, dilations = var_1769_dilations_0, groups = var_1769_groups_0, pad = var_1769_pad_0, pad_type = var_1769_pad_type_0, strides = var_1769_strides_0, weight = layers_8_fc1_inlier_module_weight_to_fp16_palettized, x = input_67_cast_fp16)[name = tensor<string, []>("op_1769_cast_fp16")]; |
| 1197 | tensor<string, []> var_1775_pad_type_0 = const()[name = tensor<string, []>("op_1775_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1198 | tensor<int32, [2]> var_1775_strides_0 = const()[name = tensor<string, []>("op_1775_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1199 | tensor<int32, [4]> var_1775_pad_0 = const()[name = tensor<string, []>("op_1775_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1200 | tensor<int32, [2]> var_1775_dilations_0 = const()[name = tensor<string, []>("op_1775_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1201 | tensor<int32, []> var_1775_groups_0 = const()[name = tensor<string, []>("op_1775_groups_0"), val = tensor<int32, []>(1)]; |
| 1202 | tensor<fp16, [3072, 768, 1, 1]> layers_8_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46641984))), name = tensor<string, []>("layers_8_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [15537]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46610816))), shape = tensor<uint32, [4]>([3072, 768, 1, 1])]; |
| 1203 | tensor<fp16, [1, 3072, 1, 1500]> var_1775_cast_fp16 = conv(dilations = var_1775_dilations_0, groups = var_1775_groups_0, pad = var_1775_pad_0, pad_type = var_1775_pad_type_0, strides = var_1775_strides_0, weight = layers_8_fc1_outlier_module_weight_to_fp16_sparsified, x = input_67_cast_fp16)[name = tensor<string, []>("op_1775_cast_fp16")]; |
| 1204 | tensor<fp16, [1, 3072, 1, 1500]> input_69_cast_fp16 = add(x = var_1769_cast_fp16, y = var_1775_cast_fp16)[name = tensor<string, []>("input_69_cast_fp16")]; |
| 1205 | tensor<string, []> input_71_mode_0 = const()[name = tensor<string, []>("input_71_mode_0"), val = tensor<string, []>("EXACT")]; |
| 1206 | tensor<fp16, [1, 3072, 1, 1500]> input_71_cast_fp16 = gelu(mode = input_71_mode_0, x = input_69_cast_fp16)[name = tensor<string, []>("input_71_cast_fp16")]; |
| 1207 | tensor<string, []> var_1786_pad_type_0 = const()[name = tensor<string, []>("op_1786_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1208 | tensor<int32, [2]> var_1786_strides_0 = const()[name = tensor<string, []>("op_1786_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1209 | tensor<int32, [4]> var_1786_pad_0 = const()[name = tensor<string, []>("op_1786_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1210 | tensor<int32, [2]> var_1786_dilations_0 = const()[name = tensor<string, []>("op_1786_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1211 | tensor<int32, []> var_1786_groups_0 = const()[name = tensor<string, []>("op_1786_groups_0"), val = tensor<int32, []>(1)]; |
| 1212 | tensor<fp16, [768, 3072, 1, 1]> layers_8_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1179648]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46936960))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48116672))), name = tensor<string, []>("layers_8_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 3072, 1, 1])]; |
| 1213 | tensor<fp16, [768]> layers_8_fc2_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_8_fc2_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48116800)))]; |
| 1214 | tensor<fp16, [1, 768, 1, 1500]> var_1786_cast_fp16 = conv(bias = layers_8_fc2_inlier_module_bias_to_fp16, dilations = var_1786_dilations_0, groups = var_1786_groups_0, pad = var_1786_pad_0, pad_type = var_1786_pad_type_0, strides = var_1786_strides_0, weight = layers_8_fc2_inlier_module_weight_to_fp16_palettized, x = input_71_cast_fp16)[name = tensor<string, []>("op_1786_cast_fp16")]; |
| 1215 | tensor<string, []> var_1792_pad_type_0 = const()[name = tensor<string, []>("op_1792_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1216 | tensor<int32, [2]> var_1792_strides_0 = const()[name = tensor<string, []>("op_1792_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1217 | tensor<int32, [4]> var_1792_pad_0 = const()[name = tensor<string, []>("op_1792_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1218 | tensor<int32, [2]> var_1792_dilations_0 = const()[name = tensor<string, []>("op_1792_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1219 | tensor<int32, []> var_1792_groups_0 = const()[name = tensor<string, []>("op_1792_groups_0"), val = tensor<int32, []>(1)]; |
| 1220 | tensor<fp16, [768, 3072, 1, 1]> layers_8_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48148992))), name = tensor<string, []>("layers_8_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [15247]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48118400))), shape = tensor<uint32, [4]>([768, 3072, 1, 1])]; |
| 1221 | tensor<fp16, [1, 768, 1, 1500]> var_1792_cast_fp16 = conv(dilations = var_1792_dilations_0, groups = var_1792_groups_0, pad = var_1792_pad_0, pad_type = var_1792_pad_type_0, strides = var_1792_strides_0, weight = layers_8_fc2_outlier_module_weight_to_fp16_sparsified, x = input_71_cast_fp16)[name = tensor<string, []>("op_1792_cast_fp16")]; |
| 1222 | tensor<fp16, [1, 768, 1, 1500]> hidden_states_21_cast_fp16 = add(x = var_1786_cast_fp16, y = var_1792_cast_fp16)[name = tensor<string, []>("hidden_states_21_cast_fp16")]; |
| 1223 | tensor<fp16, [1, 768, 1, 1500]> inputs_37_cast_fp16 = add(x = inputs_35_cast_fp16, y = hidden_states_21_cast_fp16)[name = tensor<string, []>("inputs_37_cast_fp16")]; |
| 1224 | tensor<int32, []> var_1798 = const()[name = tensor<string, []>("op_1798"), val = tensor<int32, []>(3)]; |
| 1225 | tensor<int32, [1]> out_37_axes_0 = const()[name = tensor<string, []>("out_37_axes_0"), val = tensor<int32, [1]>([1])]; |
| 1226 | tensor<fp16, []> var_1820_to_fp16 = const()[name = tensor<string, []>("op_1820_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 1227 | tensor<fp16, [1, 768, 1, 1500]> out_37_cast_fp16 = layer_norm(axes = out_37_axes_0, epsilon = var_1820_to_fp16, x = inputs_37_cast_fp16)[name = tensor<string, []>("out_37_cast_fp16")]; |
| 1228 | tensor<fp16, [768]> obj_37_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_37_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48443968)))]; |
| 1229 | tensor<fp16, [768]> obj_37_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_37_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48445568)))]; |
| 1230 | tensor<fp16, []> obj_37_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_37_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 1231 | tensor<fp16, [1, 768, 1, 1500]> obj_37_cast_fp16 = batch_norm(beta = obj_37_beta_0_to_fp16, epsilon = obj_37_epsilon_0_to_fp16, gamma = obj_37_gamma_0_to_fp16, mean = var_57_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_37_cast_fp16)[name = tensor<string, []>("obj_37_cast_fp16")]; |
| 1232 | tensor<string, []> var_1842_pad_type_0 = const()[name = tensor<string, []>("op_1842_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1233 | tensor<int32, [2]> var_1842_strides_0 = const()[name = tensor<string, []>("op_1842_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1234 | tensor<int32, [4]> var_1842_pad_0 = const()[name = tensor<string, []>("op_1842_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1235 | tensor<int32, [2]> var_1842_dilations_0 = const()[name = tensor<string, []>("op_1842_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1236 | tensor<int32, []> var_1842_groups_0 = const()[name = tensor<string, []>("op_1842_groups_0"), val = tensor<int32, []>(1)]; |
| 1237 | tensor<fp16, [768, 768, 1, 1]> layers_9_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48447168))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48742144))), name = tensor<string, []>("layers_9_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1238 | tensor<fp16, [768]> layers_9_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_9_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48742272)))]; |
| 1239 | tensor<fp16, [1, 768, 1, 1500]> var_1842_cast_fp16 = conv(bias = layers_9_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_1842_dilations_0, groups = var_1842_groups_0, pad = var_1842_pad_0, pad_type = var_1842_pad_type_0, strides = var_1842_strides_0, weight = layers_9_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_37_cast_fp16)[name = tensor<string, []>("op_1842_cast_fp16")]; |
| 1240 | tensor<string, []> var_1848_pad_type_0 = const()[name = tensor<string, []>("op_1848_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1241 | tensor<int32, [2]> var_1848_strides_0 = const()[name = tensor<string, []>("op_1848_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1242 | tensor<int32, [4]> var_1848_pad_0 = const()[name = tensor<string, []>("op_1848_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1243 | tensor<int32, [2]> var_1848_dilations_0 = const()[name = tensor<string, []>("op_1848_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1244 | tensor<int32, []> var_1848_groups_0 = const()[name = tensor<string, []>("op_1848_groups_0"), val = tensor<int32, []>(1)]; |
| 1245 | tensor<fp16, [768, 768, 1, 1]> layers_9_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48750080))), name = tensor<string, []>("layers_9_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [3064]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48743872))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1246 | tensor<fp16, [1, 768, 1, 1500]> var_1848_cast_fp16 = conv(dilations = var_1848_dilations_0, groups = var_1848_groups_0, pad = var_1848_pad_0, pad_type = var_1848_pad_type_0, strides = var_1848_strides_0, weight = layers_9_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_37_cast_fp16)[name = tensor<string, []>("op_1848_cast_fp16")]; |
| 1247 | tensor<fp16, [1, 768, 1, 1500]> query_19_cast_fp16 = add(x = var_1842_cast_fp16, y = var_1848_cast_fp16)[name = tensor<string, []>("query_19_cast_fp16")]; |
| 1248 | tensor<string, []> var_1857_pad_type_0 = const()[name = tensor<string, []>("op_1857_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1249 | tensor<int32, [2]> var_1857_strides_0 = const()[name = tensor<string, []>("op_1857_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1250 | tensor<int32, [4]> var_1857_pad_0 = const()[name = tensor<string, []>("op_1857_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1251 | tensor<int32, [2]> var_1857_dilations_0 = const()[name = tensor<string, []>("op_1857_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1252 | tensor<int32, []> var_1857_groups_0 = const()[name = tensor<string, []>("op_1857_groups_0"), val = tensor<int32, []>(1)]; |
| 1253 | tensor<fp16, [768, 768, 1, 1]> layers_9_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48823872))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49118848))), name = tensor<string, []>("layers_9_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1254 | tensor<fp16, [1, 768, 1, 1500]> var_1857_cast_fp16 = conv(dilations = var_1857_dilations_0, groups = var_1857_groups_0, pad = var_1857_pad_0, pad_type = var_1857_pad_type_0, strides = var_1857_strides_0, weight = layers_9_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_37_cast_fp16)[name = tensor<string, []>("op_1857_cast_fp16")]; |
| 1255 | tensor<string, []> var_1863_pad_type_0 = const()[name = tensor<string, []>("op_1863_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1256 | tensor<int32, [2]> var_1863_strides_0 = const()[name = tensor<string, []>("op_1863_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1257 | tensor<int32, [4]> var_1863_pad_0 = const()[name = tensor<string, []>("op_1863_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1258 | tensor<int32, [2]> var_1863_dilations_0 = const()[name = tensor<string, []>("op_1863_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1259 | tensor<int32, []> var_1863_groups_0 = const()[name = tensor<string, []>("op_1863_groups_0"), val = tensor<int32, []>(1)]; |
| 1260 | tensor<fp16, [768, 768, 1, 1]> layers_9_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49125760))), name = tensor<string, []>("layers_9_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [3358]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49118976))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1261 | tensor<fp16, [1, 768, 1, 1500]> var_1863_cast_fp16 = conv(dilations = var_1863_dilations_0, groups = var_1863_groups_0, pad = var_1863_pad_0, pad_type = var_1863_pad_type_0, strides = var_1863_strides_0, weight = layers_9_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_37_cast_fp16)[name = tensor<string, []>("op_1863_cast_fp16")]; |
| 1262 | tensor<fp16, [1, 768, 1, 1500]> key_19_cast_fp16 = add(x = var_1857_cast_fp16, y = var_1863_cast_fp16)[name = tensor<string, []>("key_19_cast_fp16")]; |
| 1263 | tensor<string, []> var_1873_pad_type_0 = const()[name = tensor<string, []>("op_1873_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1264 | tensor<int32, [2]> var_1873_strides_0 = const()[name = tensor<string, []>("op_1873_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1265 | tensor<int32, [4]> var_1873_pad_0 = const()[name = tensor<string, []>("op_1873_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1266 | tensor<int32, [2]> var_1873_dilations_0 = const()[name = tensor<string, []>("op_1873_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1267 | tensor<int32, []> var_1873_groups_0 = const()[name = tensor<string, []>("op_1873_groups_0"), val = tensor<int32, []>(1)]; |
| 1268 | tensor<fp16, [768, 768, 1, 1]> layers_9_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49199552))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49494528))), name = tensor<string, []>("layers_9_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1269 | tensor<fp16, [768]> layers_9_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_9_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49494656)))]; |
| 1270 | tensor<fp16, [1, 768, 1, 1500]> var_1873_cast_fp16 = conv(bias = layers_9_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_1873_dilations_0, groups = var_1873_groups_0, pad = var_1873_pad_0, pad_type = var_1873_pad_type_0, strides = var_1873_strides_0, weight = layers_9_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_37_cast_fp16)[name = tensor<string, []>("op_1873_cast_fp16")]; |
| 1271 | tensor<string, []> var_1879_pad_type_0 = const()[name = tensor<string, []>("op_1879_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1272 | tensor<int32, [2]> var_1879_strides_0 = const()[name = tensor<string, []>("op_1879_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1273 | tensor<int32, [4]> var_1879_pad_0 = const()[name = tensor<string, []>("op_1879_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1274 | tensor<int32, [2]> var_1879_dilations_0 = const()[name = tensor<string, []>("op_1879_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1275 | tensor<int32, []> var_1879_groups_0 = const()[name = tensor<string, []>("op_1879_groups_0"), val = tensor<int32, []>(1)]; |
| 1276 | tensor<fp16, [768, 768, 1, 1]> layers_9_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49501760))), name = tensor<string, []>("layers_9_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [2711]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49496256))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1277 | tensor<fp16, [1, 768, 1, 1500]> var_1879_cast_fp16 = conv(dilations = var_1879_dilations_0, groups = var_1879_groups_0, pad = var_1879_pad_0, pad_type = var_1879_pad_type_0, strides = var_1879_strides_0, weight = layers_9_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_37_cast_fp16)[name = tensor<string, []>("op_1879_cast_fp16")]; |
| 1278 | tensor<fp16, [1, 768, 1, 1500]> value_19_cast_fp16 = add(x = var_1873_cast_fp16, y = var_1879_cast_fp16)[name = tensor<string, []>("value_19_cast_fp16")]; |
| 1279 | tensor<int32, [4]> var_1883 = const()[name = tensor<string, []>("op_1883"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 1280 | tensor<fp16, [1, 12, 64, 1500]> mh_q_19_cast_fp16 = reshape(shape = var_1883, x = query_19_cast_fp16)[name = tensor<string, []>("mh_q_19_cast_fp16")]; |
| 1281 | tensor<fp16, []> var_1885_to_fp16 = const()[name = tensor<string, []>("op_1885_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| 1282 | tensor<fp16, [1, 12, 64, 1500]> var_1886_cast_fp16 = mul(x = mh_q_19_cast_fp16, y = var_1885_to_fp16)[name = tensor<string, []>("op_1886_cast_fp16")]; |
| 1283 | tensor<int32, [4]> var_1889 = const()[name = tensor<string, []>("op_1889"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 1284 | tensor<fp16, [1, 12, 64, 1500]> var_1890_cast_fp16 = reshape(shape = var_1889, x = key_19_cast_fp16)[name = tensor<string, []>("op_1890_cast_fp16")]; |
| 1285 | tensor<bool, []> mh_w_19_transpose_x_0 = const()[name = tensor<string, []>("mh_w_19_transpose_x_0"), val = tensor<bool, []>(true)]; |
| 1286 | tensor<bool, []> mh_w_19_transpose_y_0 = const()[name = tensor<string, []>("mh_w_19_transpose_y_0"), val = tensor<bool, []>(false)]; |
| 1287 | tensor<fp16, [1, 12, 1500, 1500]> mh_w_19_cast_fp16 = matmul(transpose_x = mh_w_19_transpose_x_0, transpose_y = mh_w_19_transpose_y_0, x = var_1886_cast_fp16, y = var_1890_cast_fp16)[name = tensor<string, []>("mh_w_19_cast_fp16")]; |
| 1288 | tensor<fp16, [1, 12, 1500, 1500]> var_1893_cast_fp16 = softmax(axis = var_1798, x = mh_w_19_cast_fp16)[name = tensor<string, []>("op_1893_cast_fp16")]; |
| 1289 | tensor<int32, [4]> var_1894 = const()[name = tensor<string, []>("op_1894"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 1290 | tensor<fp16, [1, 12, 64, 1500]> var_1895_cast_fp16 = reshape(shape = var_1894, x = value_19_cast_fp16)[name = tensor<string, []>("op_1895_cast_fp16")]; |
| 1291 | tensor<bool, []> attn_19_transpose_x_0 = const()[name = tensor<string, []>("attn_19_transpose_x_0"), val = tensor<bool, []>(false)]; |
| 1292 | tensor<bool, []> attn_19_transpose_y_0 = const()[name = tensor<string, []>("attn_19_transpose_y_0"), val = tensor<bool, []>(true)]; |
| 1293 | tensor<fp16, [1, 12, 64, 1500]> attn_19_cast_fp16 = matmul(transpose_x = attn_19_transpose_x_0, transpose_y = attn_19_transpose_y_0, x = var_1895_cast_fp16, y = var_1893_cast_fp16)[name = tensor<string, []>("attn_19_cast_fp16")]; |
| 1294 | tensor<int32, [4]> var_1898 = const()[name = tensor<string, []>("op_1898"), val = tensor<int32, [4]>([1, 768, 1, 1500])]; |
| 1295 | tensor<fp16, [1, 768, 1, 1500]> input_73_cast_fp16 = reshape(shape = var_1898, x = attn_19_cast_fp16)[name = tensor<string, []>("input_73_cast_fp16")]; |
| 1296 | tensor<string, []> var_1908_pad_type_0 = const()[name = tensor<string, []>("op_1908_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1297 | tensor<int32, [2]> var_1908_strides_0 = const()[name = tensor<string, []>("op_1908_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1298 | tensor<int32, [4]> var_1908_pad_0 = const()[name = tensor<string, []>("op_1908_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1299 | tensor<int32, [2]> var_1908_dilations_0 = const()[name = tensor<string, []>("op_1908_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1300 | tensor<int32, []> var_1908_groups_0 = const()[name = tensor<string, []>("op_1908_groups_0"), val = tensor<int32, []>(1)]; |
| 1301 | tensor<fp16, [768, 768, 1, 1]> layers_9_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49575552))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49870528))), name = tensor<string, []>("layers_9_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1302 | tensor<fp16, [768]> layers_9_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_9_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49870656)))]; |
| 1303 | tensor<fp16, [1, 768, 1, 1500]> var_1908_cast_fp16 = conv(bias = layers_9_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_1908_dilations_0, groups = var_1908_groups_0, pad = var_1908_pad_0, pad_type = var_1908_pad_type_0, strides = var_1908_strides_0, weight = layers_9_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_73_cast_fp16)[name = tensor<string, []>("op_1908_cast_fp16")]; |
| 1304 | tensor<string, []> var_1914_pad_type_0 = const()[name = tensor<string, []>("op_1914_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1305 | tensor<int32, [2]> var_1914_strides_0 = const()[name = tensor<string, []>("op_1914_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1306 | tensor<int32, [4]> var_1914_pad_0 = const()[name = tensor<string, []>("op_1914_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1307 | tensor<int32, [2]> var_1914_dilations_0 = const()[name = tensor<string, []>("op_1914_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1308 | tensor<int32, []> var_1914_groups_0 = const()[name = tensor<string, []>("op_1914_groups_0"), val = tensor<int32, []>(1)]; |
| 1309 | tensor<fp16, [768, 768, 1, 1]> layers_9_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49877760))), name = tensor<string, []>("layers_9_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [2712]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49872256))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1310 | tensor<fp16, [1, 768, 1, 1500]> var_1914_cast_fp16 = conv(dilations = var_1914_dilations_0, groups = var_1914_groups_0, pad = var_1914_pad_0, pad_type = var_1914_pad_type_0, strides = var_1914_strides_0, weight = layers_9_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_73_cast_fp16)[name = tensor<string, []>("op_1914_cast_fp16")]; |
| 1311 | tensor<fp16, [1, 768, 1, 1500]> obj_39_cast_fp16 = add(x = var_1908_cast_fp16, y = var_1914_cast_fp16)[name = tensor<string, []>("obj_39_cast_fp16")]; |
| 1312 | tensor<fp16, [1, 768, 1, 1500]> inputs_39_cast_fp16 = add(x = inputs_37_cast_fp16, y = obj_39_cast_fp16)[name = tensor<string, []>("inputs_39_cast_fp16")]; |
| 1313 | tensor<int32, [1]> out_39_axes_0 = const()[name = tensor<string, []>("out_39_axes_0"), val = tensor<int32, [1]>([1])]; |
| 1314 | tensor<fp16, []> var_1925_to_fp16 = const()[name = tensor<string, []>("op_1925_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 1315 | tensor<fp16, [1, 768, 1, 1500]> out_39_cast_fp16 = layer_norm(axes = out_39_axes_0, epsilon = var_1925_to_fp16, x = inputs_39_cast_fp16)[name = tensor<string, []>("out_39_cast_fp16")]; |
| 1316 | tensor<fp16, [768]> input_75_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_75_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49951552)))]; |
| 1317 | tensor<fp16, [768]> input_75_beta_0_to_fp16 = const()[name = tensor<string, []>("input_75_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49953152)))]; |
| 1318 | tensor<fp16, []> input_75_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_75_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 1319 | tensor<fp16, [1, 768, 1, 1500]> input_75_cast_fp16 = batch_norm(beta = input_75_beta_0_to_fp16, epsilon = input_75_epsilon_0_to_fp16, gamma = input_75_gamma_0_to_fp16, mean = var_57_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_39_cast_fp16)[name = tensor<string, []>("input_75_cast_fp16")]; |
| 1320 | tensor<string, []> var_1943_pad_type_0 = const()[name = tensor<string, []>("op_1943_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1321 | tensor<int32, [2]> var_1943_strides_0 = const()[name = tensor<string, []>("op_1943_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1322 | tensor<int32, [4]> var_1943_pad_0 = const()[name = tensor<string, []>("op_1943_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1323 | tensor<int32, [2]> var_1943_dilations_0 = const()[name = tensor<string, []>("op_1943_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1324 | tensor<int32, []> var_1943_groups_0 = const()[name = tensor<string, []>("op_1943_groups_0"), val = tensor<int32, []>(1)]; |
| 1325 | tensor<fp16, [3072, 768, 1, 1]> layers_9_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1179648]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49954752))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51134464))), name = tensor<string, []>("layers_9_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([3072, 768, 1, 1])]; |
| 1326 | tensor<fp16, [3072]> layers_9_fc1_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_9_fc1_inlier_module_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51134592)))]; |
| 1327 | tensor<fp16, [1, 3072, 1, 1500]> var_1943_cast_fp16 = conv(bias = layers_9_fc1_inlier_module_bias_to_fp16, dilations = var_1943_dilations_0, groups = var_1943_groups_0, pad = var_1943_pad_0, pad_type = var_1943_pad_type_0, strides = var_1943_strides_0, weight = layers_9_fc1_inlier_module_weight_to_fp16_palettized, x = input_75_cast_fp16)[name = tensor<string, []>("op_1943_cast_fp16")]; |
| 1328 | tensor<string, []> var_1949_pad_type_0 = const()[name = tensor<string, []>("op_1949_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1329 | tensor<int32, [2]> var_1949_strides_0 = const()[name = tensor<string, []>("op_1949_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1330 | tensor<int32, [4]> var_1949_pad_0 = const()[name = tensor<string, []>("op_1949_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1331 | tensor<int32, [2]> var_1949_dilations_0 = const()[name = tensor<string, []>("op_1949_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1332 | tensor<int32, []> var_1949_groups_0 = const()[name = tensor<string, []>("op_1949_groups_0"), val = tensor<int32, []>(1)]; |
| 1333 | tensor<fp16, [3072, 768, 1, 1]> layers_9_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51171648))), name = tensor<string, []>("layers_9_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [15382]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51140800))), shape = tensor<uint32, [4]>([3072, 768, 1, 1])]; |
| 1334 | tensor<fp16, [1, 3072, 1, 1500]> var_1949_cast_fp16 = conv(dilations = var_1949_dilations_0, groups = var_1949_groups_0, pad = var_1949_pad_0, pad_type = var_1949_pad_type_0, strides = var_1949_strides_0, weight = layers_9_fc1_outlier_module_weight_to_fp16_sparsified, x = input_75_cast_fp16)[name = tensor<string, []>("op_1949_cast_fp16")]; |
| 1335 | tensor<fp16, [1, 3072, 1, 1500]> input_77_cast_fp16 = add(x = var_1943_cast_fp16, y = var_1949_cast_fp16)[name = tensor<string, []>("input_77_cast_fp16")]; |
| 1336 | tensor<string, []> input_79_mode_0 = const()[name = tensor<string, []>("input_79_mode_0"), val = tensor<string, []>("EXACT")]; |
| 1337 | tensor<fp16, [1, 3072, 1, 1500]> input_79_cast_fp16 = gelu(mode = input_79_mode_0, x = input_77_cast_fp16)[name = tensor<string, []>("input_79_cast_fp16")]; |
| 1338 | tensor<string, []> var_1960_pad_type_0 = const()[name = tensor<string, []>("op_1960_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1339 | tensor<int32, [2]> var_1960_strides_0 = const()[name = tensor<string, []>("op_1960_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1340 | tensor<int32, [4]> var_1960_pad_0 = const()[name = tensor<string, []>("op_1960_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1341 | tensor<int32, [2]> var_1960_dilations_0 = const()[name = tensor<string, []>("op_1960_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1342 | tensor<int32, []> var_1960_groups_0 = const()[name = tensor<string, []>("op_1960_groups_0"), val = tensor<int32, []>(1)]; |
| 1343 | tensor<fp16, [768, 3072, 1, 1]> layers_9_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1179648]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51466624))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52646336))), name = tensor<string, []>("layers_9_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 3072, 1, 1])]; |
| 1344 | tensor<fp16, [768]> layers_9_fc2_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_9_fc2_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52646464)))]; |
| 1345 | tensor<fp16, [1, 768, 1, 1500]> var_1960_cast_fp16 = conv(bias = layers_9_fc2_inlier_module_bias_to_fp16, dilations = var_1960_dilations_0, groups = var_1960_groups_0, pad = var_1960_pad_0, pad_type = var_1960_pad_type_0, strides = var_1960_strides_0, weight = layers_9_fc2_inlier_module_weight_to_fp16_palettized, x = input_79_cast_fp16)[name = tensor<string, []>("op_1960_cast_fp16")]; |
| 1346 | tensor<string, []> var_1966_pad_type_0 = const()[name = tensor<string, []>("op_1966_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1347 | tensor<int32, [2]> var_1966_strides_0 = const()[name = tensor<string, []>("op_1966_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1348 | tensor<int32, [4]> var_1966_pad_0 = const()[name = tensor<string, []>("op_1966_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1349 | tensor<int32, [2]> var_1966_dilations_0 = const()[name = tensor<string, []>("op_1966_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1350 | tensor<int32, []> var_1966_groups_0 = const()[name = tensor<string, []>("op_1966_groups_0"), val = tensor<int32, []>(1)]; |
| 1351 | tensor<fp16, [768, 3072, 1, 1]> layers_9_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52680576))), name = tensor<string, []>("layers_9_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [16208]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52648064))), shape = tensor<uint32, [4]>([768, 3072, 1, 1])]; |
| 1352 | tensor<fp16, [1, 768, 1, 1500]> var_1966_cast_fp16 = conv(dilations = var_1966_dilations_0, groups = var_1966_groups_0, pad = var_1966_pad_0, pad_type = var_1966_pad_type_0, strides = var_1966_strides_0, weight = layers_9_fc2_outlier_module_weight_to_fp16_sparsified, x = input_79_cast_fp16)[name = tensor<string, []>("op_1966_cast_fp16")]; |
| 1353 | tensor<fp16, [1, 768, 1, 1500]> hidden_states_23_cast_fp16 = add(x = var_1960_cast_fp16, y = var_1966_cast_fp16)[name = tensor<string, []>("hidden_states_23_cast_fp16")]; |
| 1354 | tensor<fp16, [1, 768, 1, 1500]> inputs_41_cast_fp16 = add(x = inputs_39_cast_fp16, y = hidden_states_23_cast_fp16)[name = tensor<string, []>("inputs_41_cast_fp16")]; |
| 1355 | tensor<int32, []> var_1972 = const()[name = tensor<string, []>("op_1972"), val = tensor<int32, []>(3)]; |
| 1356 | tensor<int32, [1]> out_41_axes_0 = const()[name = tensor<string, []>("out_41_axes_0"), val = tensor<int32, [1]>([1])]; |
| 1357 | tensor<fp16, []> var_1994_to_fp16 = const()[name = tensor<string, []>("op_1994_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 1358 | tensor<fp16, [1, 768, 1, 1500]> out_41_cast_fp16 = layer_norm(axes = out_41_axes_0, epsilon = var_1994_to_fp16, x = inputs_41_cast_fp16)[name = tensor<string, []>("out_41_cast_fp16")]; |
| 1359 | tensor<fp16, [768]> obj_41_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_41_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52975552)))]; |
| 1360 | tensor<fp16, [768]> obj_41_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_41_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52977152)))]; |
| 1361 | tensor<fp16, []> obj_41_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_41_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 1362 | tensor<fp16, [1, 768, 1, 1500]> obj_41_cast_fp16 = batch_norm(beta = obj_41_beta_0_to_fp16, epsilon = obj_41_epsilon_0_to_fp16, gamma = obj_41_gamma_0_to_fp16, mean = var_57_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_41_cast_fp16)[name = tensor<string, []>("obj_41_cast_fp16")]; |
| 1363 | tensor<string, []> var_2016_pad_type_0 = const()[name = tensor<string, []>("op_2016_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1364 | tensor<int32, [2]> var_2016_strides_0 = const()[name = tensor<string, []>("op_2016_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1365 | tensor<int32, [4]> var_2016_pad_0 = const()[name = tensor<string, []>("op_2016_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1366 | tensor<int32, [2]> var_2016_dilations_0 = const()[name = tensor<string, []>("op_2016_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1367 | tensor<int32, []> var_2016_groups_0 = const()[name = tensor<string, []>("op_2016_groups_0"), val = tensor<int32, []>(1)]; |
| 1368 | tensor<fp16, [768, 768, 1, 1]> layers_10_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52978752))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53273728))), name = tensor<string, []>("layers_10_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1369 | tensor<fp16, [768]> layers_10_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_10_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53273856)))]; |
| 1370 | tensor<fp16, [1, 768, 1, 1500]> var_2016_cast_fp16 = conv(bias = layers_10_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_2016_dilations_0, groups = var_2016_groups_0, pad = var_2016_pad_0, pad_type = var_2016_pad_type_0, strides = var_2016_strides_0, weight = layers_10_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_41_cast_fp16)[name = tensor<string, []>("op_2016_cast_fp16")]; |
| 1371 | tensor<string, []> var_2022_pad_type_0 = const()[name = tensor<string, []>("op_2022_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1372 | tensor<int32, [2]> var_2022_strides_0 = const()[name = tensor<string, []>("op_2022_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1373 | tensor<int32, [4]> var_2022_pad_0 = const()[name = tensor<string, []>("op_2022_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1374 | tensor<int32, [2]> var_2022_dilations_0 = const()[name = tensor<string, []>("op_2022_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1375 | tensor<int32, []> var_2022_groups_0 = const()[name = tensor<string, []>("op_2022_groups_0"), val = tensor<int32, []>(1)]; |
| 1376 | tensor<fp16, [768, 768, 1, 1]> layers_10_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53281536))), name = tensor<string, []>("layers_10_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [3007]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53275456))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1377 | tensor<fp16, [1, 768, 1, 1500]> var_2022_cast_fp16 = conv(dilations = var_2022_dilations_0, groups = var_2022_groups_0, pad = var_2022_pad_0, pad_type = var_2022_pad_type_0, strides = var_2022_strides_0, weight = layers_10_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_41_cast_fp16)[name = tensor<string, []>("op_2022_cast_fp16")]; |
| 1378 | tensor<fp16, [1, 768, 1, 1500]> query_21_cast_fp16 = add(x = var_2016_cast_fp16, y = var_2022_cast_fp16)[name = tensor<string, []>("query_21_cast_fp16")]; |
| 1379 | tensor<string, []> var_2031_pad_type_0 = const()[name = tensor<string, []>("op_2031_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1380 | tensor<int32, [2]> var_2031_strides_0 = const()[name = tensor<string, []>("op_2031_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1381 | tensor<int32, [4]> var_2031_pad_0 = const()[name = tensor<string, []>("op_2031_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1382 | tensor<int32, [2]> var_2031_dilations_0 = const()[name = tensor<string, []>("op_2031_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1383 | tensor<int32, []> var_2031_groups_0 = const()[name = tensor<string, []>("op_2031_groups_0"), val = tensor<int32, []>(1)]; |
| 1384 | tensor<fp16, [768, 768, 1, 1]> layers_10_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53355328))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53650304))), name = tensor<string, []>("layers_10_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1385 | tensor<fp16, [1, 768, 1, 1500]> var_2031_cast_fp16 = conv(dilations = var_2031_dilations_0, groups = var_2031_groups_0, pad = var_2031_pad_0, pad_type = var_2031_pad_type_0, strides = var_2031_strides_0, weight = layers_10_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_41_cast_fp16)[name = tensor<string, []>("op_2031_cast_fp16")]; |
| 1386 | tensor<string, []> var_2037_pad_type_0 = const()[name = tensor<string, []>("op_2037_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1387 | tensor<int32, [2]> var_2037_strides_0 = const()[name = tensor<string, []>("op_2037_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1388 | tensor<int32, [4]> var_2037_pad_0 = const()[name = tensor<string, []>("op_2037_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1389 | tensor<int32, [2]> var_2037_dilations_0 = const()[name = tensor<string, []>("op_2037_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1390 | tensor<int32, []> var_2037_groups_0 = const()[name = tensor<string, []>("op_2037_groups_0"), val = tensor<int32, []>(1)]; |
| 1391 | tensor<fp16, [768, 768, 1, 1]> layers_10_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53657216))), name = tensor<string, []>("layers_10_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [3358]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53650432))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1392 | tensor<fp16, [1, 768, 1, 1500]> var_2037_cast_fp16 = conv(dilations = var_2037_dilations_0, groups = var_2037_groups_0, pad = var_2037_pad_0, pad_type = var_2037_pad_type_0, strides = var_2037_strides_0, weight = layers_10_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_41_cast_fp16)[name = tensor<string, []>("op_2037_cast_fp16")]; |
| 1393 | tensor<fp16, [1, 768, 1, 1500]> key_21_cast_fp16 = add(x = var_2031_cast_fp16, y = var_2037_cast_fp16)[name = tensor<string, []>("key_21_cast_fp16")]; |
| 1394 | tensor<string, []> var_2047_pad_type_0 = const()[name = tensor<string, []>("op_2047_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1395 | tensor<int32, [2]> var_2047_strides_0 = const()[name = tensor<string, []>("op_2047_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1396 | tensor<int32, [4]> var_2047_pad_0 = const()[name = tensor<string, []>("op_2047_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1397 | tensor<int32, [2]> var_2047_dilations_0 = const()[name = tensor<string, []>("op_2047_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1398 | tensor<int32, []> var_2047_groups_0 = const()[name = tensor<string, []>("op_2047_groups_0"), val = tensor<int32, []>(1)]; |
| 1399 | tensor<fp16, [768, 768, 1, 1]> layers_10_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53731008))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54025984))), name = tensor<string, []>("layers_10_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1400 | tensor<fp16, [768]> layers_10_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_10_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54026112)))]; |
| 1401 | tensor<fp16, [1, 768, 1, 1500]> var_2047_cast_fp16 = conv(bias = layers_10_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_2047_dilations_0, groups = var_2047_groups_0, pad = var_2047_pad_0, pad_type = var_2047_pad_type_0, strides = var_2047_strides_0, weight = layers_10_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_41_cast_fp16)[name = tensor<string, []>("op_2047_cast_fp16")]; |
| 1402 | tensor<string, []> var_2053_pad_type_0 = const()[name = tensor<string, []>("op_2053_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1403 | tensor<int32, [2]> var_2053_strides_0 = const()[name = tensor<string, []>("op_2053_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1404 | tensor<int32, [4]> var_2053_pad_0 = const()[name = tensor<string, []>("op_2053_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1405 | tensor<int32, [2]> var_2053_dilations_0 = const()[name = tensor<string, []>("op_2053_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1406 | tensor<int32, []> var_2053_groups_0 = const()[name = tensor<string, []>("op_2053_groups_0"), val = tensor<int32, []>(1)]; |
| 1407 | tensor<fp16, [768, 768, 1, 1]> layers_10_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54033152))), name = tensor<string, []>("layers_10_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [2685]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54027712))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1408 | tensor<fp16, [1, 768, 1, 1500]> var_2053_cast_fp16 = conv(dilations = var_2053_dilations_0, groups = var_2053_groups_0, pad = var_2053_pad_0, pad_type = var_2053_pad_type_0, strides = var_2053_strides_0, weight = layers_10_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_41_cast_fp16)[name = tensor<string, []>("op_2053_cast_fp16")]; |
| 1409 | tensor<fp16, [1, 768, 1, 1500]> value_21_cast_fp16 = add(x = var_2047_cast_fp16, y = var_2053_cast_fp16)[name = tensor<string, []>("value_21_cast_fp16")]; |
| 1410 | tensor<int32, [4]> var_2057 = const()[name = tensor<string, []>("op_2057"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 1411 | tensor<fp16, [1, 12, 64, 1500]> mh_q_21_cast_fp16 = reshape(shape = var_2057, x = query_21_cast_fp16)[name = tensor<string, []>("mh_q_21_cast_fp16")]; |
| 1412 | tensor<fp16, []> var_2059_to_fp16 = const()[name = tensor<string, []>("op_2059_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| 1413 | tensor<fp16, [1, 12, 64, 1500]> var_2060_cast_fp16 = mul(x = mh_q_21_cast_fp16, y = var_2059_to_fp16)[name = tensor<string, []>("op_2060_cast_fp16")]; |
| 1414 | tensor<int32, [4]> var_2063 = const()[name = tensor<string, []>("op_2063"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 1415 | tensor<fp16, [1, 12, 64, 1500]> var_2064_cast_fp16 = reshape(shape = var_2063, x = key_21_cast_fp16)[name = tensor<string, []>("op_2064_cast_fp16")]; |
| 1416 | tensor<bool, []> mh_w_21_transpose_x_0 = const()[name = tensor<string, []>("mh_w_21_transpose_x_0"), val = tensor<bool, []>(true)]; |
| 1417 | tensor<bool, []> mh_w_21_transpose_y_0 = const()[name = tensor<string, []>("mh_w_21_transpose_y_0"), val = tensor<bool, []>(false)]; |
| 1418 | tensor<fp16, [1, 12, 1500, 1500]> mh_w_21_cast_fp16 = matmul(transpose_x = mh_w_21_transpose_x_0, transpose_y = mh_w_21_transpose_y_0, x = var_2060_cast_fp16, y = var_2064_cast_fp16)[name = tensor<string, []>("mh_w_21_cast_fp16")]; |
| 1419 | tensor<fp16, [1, 12, 1500, 1500]> var_2067_cast_fp16 = softmax(axis = var_1972, x = mh_w_21_cast_fp16)[name = tensor<string, []>("op_2067_cast_fp16")]; |
| 1420 | tensor<int32, [4]> var_2068 = const()[name = tensor<string, []>("op_2068"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 1421 | tensor<fp16, [1, 12, 64, 1500]> var_2069_cast_fp16 = reshape(shape = var_2068, x = value_21_cast_fp16)[name = tensor<string, []>("op_2069_cast_fp16")]; |
| 1422 | tensor<bool, []> attn_21_transpose_x_0 = const()[name = tensor<string, []>("attn_21_transpose_x_0"), val = tensor<bool, []>(false)]; |
| 1423 | tensor<bool, []> attn_21_transpose_y_0 = const()[name = tensor<string, []>("attn_21_transpose_y_0"), val = tensor<bool, []>(true)]; |
| 1424 | tensor<fp16, [1, 12, 64, 1500]> attn_21_cast_fp16 = matmul(transpose_x = attn_21_transpose_x_0, transpose_y = attn_21_transpose_y_0, x = var_2069_cast_fp16, y = var_2067_cast_fp16)[name = tensor<string, []>("attn_21_cast_fp16")]; |
| 1425 | tensor<int32, [4]> var_2072 = const()[name = tensor<string, []>("op_2072"), val = tensor<int32, [4]>([1, 768, 1, 1500])]; |
| 1426 | tensor<fp16, [1, 768, 1, 1500]> input_81_cast_fp16 = reshape(shape = var_2072, x = attn_21_cast_fp16)[name = tensor<string, []>("input_81_cast_fp16")]; |
| 1427 | tensor<string, []> var_2082_pad_type_0 = const()[name = tensor<string, []>("op_2082_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1428 | tensor<int32, [2]> var_2082_strides_0 = const()[name = tensor<string, []>("op_2082_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1429 | tensor<int32, [4]> var_2082_pad_0 = const()[name = tensor<string, []>("op_2082_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1430 | tensor<int32, [2]> var_2082_dilations_0 = const()[name = tensor<string, []>("op_2082_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1431 | tensor<int32, []> var_2082_groups_0 = const()[name = tensor<string, []>("op_2082_groups_0"), val = tensor<int32, []>(1)]; |
| 1432 | tensor<fp16, [768, 768, 1, 1]> layers_10_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54106944))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54401920))), name = tensor<string, []>("layers_10_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1433 | tensor<fp16, [768]> layers_10_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_10_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54402048)))]; |
| 1434 | tensor<fp16, [1, 768, 1, 1500]> var_2082_cast_fp16 = conv(bias = layers_10_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_2082_dilations_0, groups = var_2082_groups_0, pad = var_2082_pad_0, pad_type = var_2082_pad_type_0, strides = var_2082_strides_0, weight = layers_10_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_81_cast_fp16)[name = tensor<string, []>("op_2082_cast_fp16")]; |
| 1435 | tensor<string, []> var_2088_pad_type_0 = const()[name = tensor<string, []>("op_2088_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1436 | tensor<int32, [2]> var_2088_strides_0 = const()[name = tensor<string, []>("op_2088_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1437 | tensor<int32, [4]> var_2088_pad_0 = const()[name = tensor<string, []>("op_2088_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1438 | tensor<int32, [2]> var_2088_dilations_0 = const()[name = tensor<string, []>("op_2088_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1439 | tensor<int32, []> var_2088_groups_0 = const()[name = tensor<string, []>("op_2088_groups_0"), val = tensor<int32, []>(1)]; |
| 1440 | tensor<fp16, [768, 768, 1, 1]> layers_10_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54409472))), name = tensor<string, []>("layers_10_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [2866]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54403648))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1441 | tensor<fp16, [1, 768, 1, 1500]> var_2088_cast_fp16 = conv(dilations = var_2088_dilations_0, groups = var_2088_groups_0, pad = var_2088_pad_0, pad_type = var_2088_pad_type_0, strides = var_2088_strides_0, weight = layers_10_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_81_cast_fp16)[name = tensor<string, []>("op_2088_cast_fp16")]; |
| 1442 | tensor<fp16, [1, 768, 1, 1500]> obj_43_cast_fp16 = add(x = var_2082_cast_fp16, y = var_2088_cast_fp16)[name = tensor<string, []>("obj_43_cast_fp16")]; |
| 1443 | tensor<fp16, [1, 768, 1, 1500]> inputs_43_cast_fp16 = add(x = inputs_41_cast_fp16, y = obj_43_cast_fp16)[name = tensor<string, []>("inputs_43_cast_fp16")]; |
| 1444 | tensor<int32, [1]> out_43_axes_0 = const()[name = tensor<string, []>("out_43_axes_0"), val = tensor<int32, [1]>([1])]; |
| 1445 | tensor<fp16, []> var_2099_to_fp16 = const()[name = tensor<string, []>("op_2099_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 1446 | tensor<fp16, [1, 768, 1, 1500]> out_43_cast_fp16 = layer_norm(axes = out_43_axes_0, epsilon = var_2099_to_fp16, x = inputs_43_cast_fp16)[name = tensor<string, []>("out_43_cast_fp16")]; |
| 1447 | tensor<fp16, [768]> input_83_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_83_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54483264)))]; |
| 1448 | tensor<fp16, [768]> input_83_beta_0_to_fp16 = const()[name = tensor<string, []>("input_83_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54484864)))]; |
| 1449 | tensor<fp16, []> input_83_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_83_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 1450 | tensor<fp16, [1, 768, 1, 1500]> input_83_cast_fp16 = batch_norm(beta = input_83_beta_0_to_fp16, epsilon = input_83_epsilon_0_to_fp16, gamma = input_83_gamma_0_to_fp16, mean = var_57_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_43_cast_fp16)[name = tensor<string, []>("input_83_cast_fp16")]; |
| 1451 | tensor<string, []> var_2117_pad_type_0 = const()[name = tensor<string, []>("op_2117_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1452 | tensor<int32, [2]> var_2117_strides_0 = const()[name = tensor<string, []>("op_2117_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1453 | tensor<int32, [4]> var_2117_pad_0 = const()[name = tensor<string, []>("op_2117_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1454 | tensor<int32, [2]> var_2117_dilations_0 = const()[name = tensor<string, []>("op_2117_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1455 | tensor<int32, []> var_2117_groups_0 = const()[name = tensor<string, []>("op_2117_groups_0"), val = tensor<int32, []>(1)]; |
| 1456 | tensor<fp16, [3072, 768, 1, 1]> layers_10_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1179648]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54486464))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55666176))), name = tensor<string, []>("layers_10_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([3072, 768, 1, 1])]; |
| 1457 | tensor<fp16, [3072]> layers_10_fc1_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_10_fc1_inlier_module_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55666304)))]; |
| 1458 | tensor<fp16, [1, 3072, 1, 1500]> var_2117_cast_fp16 = conv(bias = layers_10_fc1_inlier_module_bias_to_fp16, dilations = var_2117_dilations_0, groups = var_2117_groups_0, pad = var_2117_pad_0, pad_type = var_2117_pad_type_0, strides = var_2117_strides_0, weight = layers_10_fc1_inlier_module_weight_to_fp16_palettized, x = input_83_cast_fp16)[name = tensor<string, []>("op_2117_cast_fp16")]; |
| 1459 | tensor<string, []> var_2123_pad_type_0 = const()[name = tensor<string, []>("op_2123_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1460 | tensor<int32, [2]> var_2123_strides_0 = const()[name = tensor<string, []>("op_2123_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1461 | tensor<int32, [4]> var_2123_pad_0 = const()[name = tensor<string, []>("op_2123_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1462 | tensor<int32, [2]> var_2123_dilations_0 = const()[name = tensor<string, []>("op_2123_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1463 | tensor<int32, []> var_2123_groups_0 = const()[name = tensor<string, []>("op_2123_groups_0"), val = tensor<int32, []>(1)]; |
| 1464 | tensor<fp16, [3072, 768, 1, 1]> layers_10_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55704320))), name = tensor<string, []>("layers_10_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [15858]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55672512))), shape = tensor<uint32, [4]>([3072, 768, 1, 1])]; |
| 1465 | tensor<fp16, [1, 3072, 1, 1500]> var_2123_cast_fp16 = conv(dilations = var_2123_dilations_0, groups = var_2123_groups_0, pad = var_2123_pad_0, pad_type = var_2123_pad_type_0, strides = var_2123_strides_0, weight = layers_10_fc1_outlier_module_weight_to_fp16_sparsified, x = input_83_cast_fp16)[name = tensor<string, []>("op_2123_cast_fp16")]; |
| 1466 | tensor<fp16, [1, 3072, 1, 1500]> input_85_cast_fp16 = add(x = var_2117_cast_fp16, y = var_2123_cast_fp16)[name = tensor<string, []>("input_85_cast_fp16")]; |
| 1467 | tensor<string, []> input_87_mode_0 = const()[name = tensor<string, []>("input_87_mode_0"), val = tensor<string, []>("EXACT")]; |
| 1468 | tensor<fp16, [1, 3072, 1, 1500]> input_87_cast_fp16 = gelu(mode = input_87_mode_0, x = input_85_cast_fp16)[name = tensor<string, []>("input_87_cast_fp16")]; |
| 1469 | tensor<string, []> var_2134_pad_type_0 = const()[name = tensor<string, []>("op_2134_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1470 | tensor<int32, [2]> var_2134_strides_0 = const()[name = tensor<string, []>("op_2134_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1471 | tensor<int32, [4]> var_2134_pad_0 = const()[name = tensor<string, []>("op_2134_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1472 | tensor<int32, [2]> var_2134_dilations_0 = const()[name = tensor<string, []>("op_2134_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1473 | tensor<int32, []> var_2134_groups_0 = const()[name = tensor<string, []>("op_2134_groups_0"), val = tensor<int32, []>(1)]; |
| 1474 | tensor<fp16, [768, 3072, 1, 1]> layers_10_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1179648]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55999296))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57179008))), name = tensor<string, []>("layers_10_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 3072, 1, 1])]; |
| 1475 | tensor<fp16, [768]> layers_10_fc2_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_10_fc2_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57179136)))]; |
| 1476 | tensor<fp16, [1, 768, 1, 1500]> var_2134_cast_fp16 = conv(bias = layers_10_fc2_inlier_module_bias_to_fp16, dilations = var_2134_dilations_0, groups = var_2134_groups_0, pad = var_2134_pad_0, pad_type = var_2134_pad_type_0, strides = var_2134_strides_0, weight = layers_10_fc2_inlier_module_weight_to_fp16_palettized, x = input_87_cast_fp16)[name = tensor<string, []>("op_2134_cast_fp16")]; |
| 1477 | tensor<string, []> var_2140_pad_type_0 = const()[name = tensor<string, []>("op_2140_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1478 | tensor<int32, [2]> var_2140_strides_0 = const()[name = tensor<string, []>("op_2140_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1479 | tensor<int32, [4]> var_2140_pad_0 = const()[name = tensor<string, []>("op_2140_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1480 | tensor<int32, [2]> var_2140_dilations_0 = const()[name = tensor<string, []>("op_2140_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1481 | tensor<int32, []> var_2140_groups_0 = const()[name = tensor<string, []>("op_2140_groups_0"), val = tensor<int32, []>(1)]; |
| 1482 | tensor<fp16, [768, 3072, 1, 1]> layers_10_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57215232))), name = tensor<string, []>("layers_10_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [17216]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57180736))), shape = tensor<uint32, [4]>([768, 3072, 1, 1])]; |
| 1483 | tensor<fp16, [1, 768, 1, 1500]> var_2140_cast_fp16 = conv(dilations = var_2140_dilations_0, groups = var_2140_groups_0, pad = var_2140_pad_0, pad_type = var_2140_pad_type_0, strides = var_2140_strides_0, weight = layers_10_fc2_outlier_module_weight_to_fp16_sparsified, x = input_87_cast_fp16)[name = tensor<string, []>("op_2140_cast_fp16")]; |
| 1484 | tensor<fp16, [1, 768, 1, 1500]> hidden_states_25_cast_fp16 = add(x = var_2134_cast_fp16, y = var_2140_cast_fp16)[name = tensor<string, []>("hidden_states_25_cast_fp16")]; |
| 1485 | tensor<fp16, [1, 768, 1, 1500]> inputs_45_cast_fp16 = add(x = inputs_43_cast_fp16, y = hidden_states_25_cast_fp16)[name = tensor<string, []>("inputs_45_cast_fp16")]; |
| 1486 | tensor<int32, []> var_2146 = const()[name = tensor<string, []>("op_2146"), val = tensor<int32, []>(3)]; |
| 1487 | tensor<int32, [1]> out_45_axes_0 = const()[name = tensor<string, []>("out_45_axes_0"), val = tensor<int32, [1]>([1])]; |
| 1488 | tensor<fp16, []> var_2168_to_fp16 = const()[name = tensor<string, []>("op_2168_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 1489 | tensor<fp16, [1, 768, 1, 1500]> out_45_cast_fp16 = layer_norm(axes = out_45_axes_0, epsilon = var_2168_to_fp16, x = inputs_45_cast_fp16)[name = tensor<string, []>("out_45_cast_fp16")]; |
| 1490 | tensor<fp16, [768]> obj_45_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_45_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57510208)))]; |
| 1491 | tensor<fp16, [768]> obj_45_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_45_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57511808)))]; |
| 1492 | tensor<fp16, []> obj_45_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_45_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 1493 | tensor<fp16, [1, 768, 1, 1500]> obj_45_cast_fp16 = batch_norm(beta = obj_45_beta_0_to_fp16, epsilon = obj_45_epsilon_0_to_fp16, gamma = obj_45_gamma_0_to_fp16, mean = var_57_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_45_cast_fp16)[name = tensor<string, []>("obj_45_cast_fp16")]; |
| 1494 | tensor<string, []> var_2190_pad_type_0 = const()[name = tensor<string, []>("op_2190_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1495 | tensor<int32, [2]> var_2190_strides_0 = const()[name = tensor<string, []>("op_2190_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1496 | tensor<int32, [4]> var_2190_pad_0 = const()[name = tensor<string, []>("op_2190_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1497 | tensor<int32, [2]> var_2190_dilations_0 = const()[name = tensor<string, []>("op_2190_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1498 | tensor<int32, []> var_2190_groups_0 = const()[name = tensor<string, []>("op_2190_groups_0"), val = tensor<int32, []>(1)]; |
| 1499 | tensor<fp16, [768, 768, 1, 1]> layers_11_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57513408))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57808384))), name = tensor<string, []>("layers_11_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1500 | tensor<fp16, [768]> layers_11_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_11_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57808512)))]; |
| 1501 | tensor<fp16, [1, 768, 1, 1500]> var_2190_cast_fp16 = conv(bias = layers_11_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_2190_dilations_0, groups = var_2190_groups_0, pad = var_2190_pad_0, pad_type = var_2190_pad_type_0, strides = var_2190_strides_0, weight = layers_11_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_45_cast_fp16)[name = tensor<string, []>("op_2190_cast_fp16")]; |
| 1502 | tensor<string, []> var_2196_pad_type_0 = const()[name = tensor<string, []>("op_2196_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1503 | tensor<int32, [2]> var_2196_strides_0 = const()[name = tensor<string, []>("op_2196_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1504 | tensor<int32, [4]> var_2196_pad_0 = const()[name = tensor<string, []>("op_2196_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1505 | tensor<int32, [2]> var_2196_dilations_0 = const()[name = tensor<string, []>("op_2196_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1506 | tensor<int32, []> var_2196_groups_0 = const()[name = tensor<string, []>("op_2196_groups_0"), val = tensor<int32, []>(1)]; |
| 1507 | tensor<fp16, [768, 768, 1, 1]> layers_11_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57816768))), name = tensor<string, []>("layers_11_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [3283]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57810112))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1508 | tensor<fp16, [1, 768, 1, 1500]> var_2196_cast_fp16 = conv(dilations = var_2196_dilations_0, groups = var_2196_groups_0, pad = var_2196_pad_0, pad_type = var_2196_pad_type_0, strides = var_2196_strides_0, weight = layers_11_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_45_cast_fp16)[name = tensor<string, []>("op_2196_cast_fp16")]; |
| 1509 | tensor<fp16, [1, 768, 1, 1500]> query_cast_fp16 = add(x = var_2190_cast_fp16, y = var_2196_cast_fp16)[name = tensor<string, []>("query_cast_fp16")]; |
| 1510 | tensor<string, []> var_2205_pad_type_0 = const()[name = tensor<string, []>("op_2205_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1511 | tensor<int32, [2]> var_2205_strides_0 = const()[name = tensor<string, []>("op_2205_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1512 | tensor<int32, [4]> var_2205_pad_0 = const()[name = tensor<string, []>("op_2205_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1513 | tensor<int32, [2]> var_2205_dilations_0 = const()[name = tensor<string, []>("op_2205_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1514 | tensor<int32, []> var_2205_groups_0 = const()[name = tensor<string, []>("op_2205_groups_0"), val = tensor<int32, []>(1)]; |
| 1515 | tensor<fp16, [768, 768, 1, 1]> layers_11_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57890560))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58185536))), name = tensor<string, []>("layers_11_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1516 | tensor<fp16, [1, 768, 1, 1500]> var_2205_cast_fp16 = conv(dilations = var_2205_dilations_0, groups = var_2205_groups_0, pad = var_2205_pad_0, pad_type = var_2205_pad_type_0, strides = var_2205_strides_0, weight = layers_11_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_45_cast_fp16)[name = tensor<string, []>("op_2205_cast_fp16")]; |
| 1517 | tensor<string, []> var_2211_pad_type_0 = const()[name = tensor<string, []>("op_2211_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1518 | tensor<int32, [2]> var_2211_strides_0 = const()[name = tensor<string, []>("op_2211_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1519 | tensor<int32, [4]> var_2211_pad_0 = const()[name = tensor<string, []>("op_2211_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1520 | tensor<int32, [2]> var_2211_dilations_0 = const()[name = tensor<string, []>("op_2211_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1521 | tensor<int32, []> var_2211_groups_0 = const()[name = tensor<string, []>("op_2211_groups_0"), val = tensor<int32, []>(1)]; |
| 1522 | tensor<fp16, [768, 768, 1, 1]> layers_11_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58192896))), name = tensor<string, []>("layers_11_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [3584]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58185664))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1523 | tensor<fp16, [1, 768, 1, 1500]> var_2211_cast_fp16 = conv(dilations = var_2211_dilations_0, groups = var_2211_groups_0, pad = var_2211_pad_0, pad_type = var_2211_pad_type_0, strides = var_2211_strides_0, weight = layers_11_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_45_cast_fp16)[name = tensor<string, []>("op_2211_cast_fp16")]; |
| 1524 | tensor<fp16, [1, 768, 1, 1500]> key_cast_fp16 = add(x = var_2205_cast_fp16, y = var_2211_cast_fp16)[name = tensor<string, []>("key_cast_fp16")]; |
| 1525 | tensor<string, []> var_2221_pad_type_0 = const()[name = tensor<string, []>("op_2221_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1526 | tensor<int32, [2]> var_2221_strides_0 = const()[name = tensor<string, []>("op_2221_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1527 | tensor<int32, [4]> var_2221_pad_0 = const()[name = tensor<string, []>("op_2221_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1528 | tensor<int32, [2]> var_2221_dilations_0 = const()[name = tensor<string, []>("op_2221_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1529 | tensor<int32, []> var_2221_groups_0 = const()[name = tensor<string, []>("op_2221_groups_0"), val = tensor<int32, []>(1)]; |
| 1530 | tensor<fp16, [768, 768, 1, 1]> layers_11_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58266688))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58561664))), name = tensor<string, []>("layers_11_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1531 | tensor<fp16, [768]> layers_11_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_11_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58561792)))]; |
| 1532 | tensor<fp16, [1, 768, 1, 1500]> var_2221_cast_fp16 = conv(bias = layers_11_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_2221_dilations_0, groups = var_2221_groups_0, pad = var_2221_pad_0, pad_type = var_2221_pad_type_0, strides = var_2221_strides_0, weight = layers_11_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_45_cast_fp16)[name = tensor<string, []>("op_2221_cast_fp16")]; |
| 1533 | tensor<string, []> var_2227_pad_type_0 = const()[name = tensor<string, []>("op_2227_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1534 | tensor<int32, [2]> var_2227_strides_0 = const()[name = tensor<string, []>("op_2227_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1535 | tensor<int32, [4]> var_2227_pad_0 = const()[name = tensor<string, []>("op_2227_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1536 | tensor<int32, [2]> var_2227_dilations_0 = const()[name = tensor<string, []>("op_2227_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1537 | tensor<int32, []> var_2227_groups_0 = const()[name = tensor<string, []>("op_2227_groups_0"), val = tensor<int32, []>(1)]; |
| 1538 | tensor<fp16, [768, 768, 1, 1]> layers_11_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58569728))), name = tensor<string, []>("layers_11_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [3113]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58563392))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1539 | tensor<fp16, [1, 768, 1, 1500]> var_2227_cast_fp16 = conv(dilations = var_2227_dilations_0, groups = var_2227_groups_0, pad = var_2227_pad_0, pad_type = var_2227_pad_type_0, strides = var_2227_strides_0, weight = layers_11_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_45_cast_fp16)[name = tensor<string, []>("op_2227_cast_fp16")]; |
| 1540 | tensor<fp16, [1, 768, 1, 1500]> value_cast_fp16 = add(x = var_2221_cast_fp16, y = var_2227_cast_fp16)[name = tensor<string, []>("value_cast_fp16")]; |
| 1541 | tensor<int32, [4]> var_2231 = const()[name = tensor<string, []>("op_2231"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 1542 | tensor<fp16, [1, 12, 64, 1500]> mh_q_cast_fp16 = reshape(shape = var_2231, x = query_cast_fp16)[name = tensor<string, []>("mh_q_cast_fp16")]; |
| 1543 | tensor<fp16, []> var_2233_to_fp16 = const()[name = tensor<string, []>("op_2233_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| 1544 | tensor<fp16, [1, 12, 64, 1500]> var_2234_cast_fp16 = mul(x = mh_q_cast_fp16, y = var_2233_to_fp16)[name = tensor<string, []>("op_2234_cast_fp16")]; |
| 1545 | tensor<int32, [4]> var_2237 = const()[name = tensor<string, []>("op_2237"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 1546 | tensor<fp16, [1, 12, 64, 1500]> var_2238_cast_fp16 = reshape(shape = var_2237, x = key_cast_fp16)[name = tensor<string, []>("op_2238_cast_fp16")]; |
| 1547 | tensor<bool, []> mh_w_transpose_x_0 = const()[name = tensor<string, []>("mh_w_transpose_x_0"), val = tensor<bool, []>(true)]; |
| 1548 | tensor<bool, []> mh_w_transpose_y_0 = const()[name = tensor<string, []>("mh_w_transpose_y_0"), val = tensor<bool, []>(false)]; |
| 1549 | tensor<fp16, [1, 12, 1500, 1500]> mh_w_cast_fp16 = matmul(transpose_x = mh_w_transpose_x_0, transpose_y = mh_w_transpose_y_0, x = var_2234_cast_fp16, y = var_2238_cast_fp16)[name = tensor<string, []>("mh_w_cast_fp16")]; |
| 1550 | tensor<fp16, [1, 12, 1500, 1500]> var_2241_cast_fp16 = softmax(axis = var_2146, x = mh_w_cast_fp16)[name = tensor<string, []>("op_2241_cast_fp16")]; |
| 1551 | tensor<int32, [4]> var_2242 = const()[name = tensor<string, []>("op_2242"), val = tensor<int32, [4]>([1, 12, 64, 1500])]; |
| 1552 | tensor<fp16, [1, 12, 64, 1500]> var_2243_cast_fp16 = reshape(shape = var_2242, x = value_cast_fp16)[name = tensor<string, []>("op_2243_cast_fp16")]; |
| 1553 | tensor<bool, []> attn_transpose_x_0 = const()[name = tensor<string, []>("attn_transpose_x_0"), val = tensor<bool, []>(false)]; |
| 1554 | tensor<bool, []> attn_transpose_y_0 = const()[name = tensor<string, []>("attn_transpose_y_0"), val = tensor<bool, []>(true)]; |
| 1555 | tensor<fp16, [1, 12, 64, 1500]> attn_cast_fp16 = matmul(transpose_x = attn_transpose_x_0, transpose_y = attn_transpose_y_0, x = var_2243_cast_fp16, y = var_2241_cast_fp16)[name = tensor<string, []>("attn_cast_fp16")]; |
| 1556 | tensor<int32, [4]> var_2246 = const()[name = tensor<string, []>("op_2246"), val = tensor<int32, [4]>([1, 768, 1, 1500])]; |
| 1557 | tensor<fp16, [1, 768, 1, 1500]> input_89_cast_fp16 = reshape(shape = var_2246, x = attn_cast_fp16)[name = tensor<string, []>("input_89_cast_fp16")]; |
| 1558 | tensor<string, []> var_2256_pad_type_0 = const()[name = tensor<string, []>("op_2256_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1559 | tensor<int32, [2]> var_2256_strides_0 = const()[name = tensor<string, []>("op_2256_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1560 | tensor<int32, [4]> var_2256_pad_0 = const()[name = tensor<string, []>("op_2256_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1561 | tensor<int32, [2]> var_2256_dilations_0 = const()[name = tensor<string, []>("op_2256_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1562 | tensor<int32, []> var_2256_groups_0 = const()[name = tensor<string, []>("op_2256_groups_0"), val = tensor<int32, []>(1)]; |
| 1563 | tensor<fp16, [768, 768, 1, 1]> layers_11_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58643520))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58938496))), name = tensor<string, []>("layers_11_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1564 | tensor<fp16, [768]> layers_11_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_11_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58938624)))]; |
| 1565 | tensor<fp16, [1, 768, 1, 1500]> var_2256_cast_fp16 = conv(bias = layers_11_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_2256_dilations_0, groups = var_2256_groups_0, pad = var_2256_pad_0, pad_type = var_2256_pad_type_0, strides = var_2256_strides_0, weight = layers_11_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_89_cast_fp16)[name = tensor<string, []>("op_2256_cast_fp16")]; |
| 1566 | tensor<string, []> var_2262_pad_type_0 = const()[name = tensor<string, []>("op_2262_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1567 | tensor<int32, [2]> var_2262_strides_0 = const()[name = tensor<string, []>("op_2262_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1568 | tensor<int32, [4]> var_2262_pad_0 = const()[name = tensor<string, []>("op_2262_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1569 | tensor<int32, [2]> var_2262_dilations_0 = const()[name = tensor<string, []>("op_2262_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1570 | tensor<int32, []> var_2262_groups_0 = const()[name = tensor<string, []>("op_2262_groups_0"), val = tensor<int32, []>(1)]; |
| 1571 | tensor<fp16, [768, 768, 1, 1]> layers_11_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [73728]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58947520))), name = tensor<string, []>("layers_11_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [3613]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58940224))), shape = tensor<uint32, [4]>([768, 768, 1, 1])]; |
| 1572 | tensor<fp16, [1, 768, 1, 1500]> var_2262_cast_fp16 = conv(dilations = var_2262_dilations_0, groups = var_2262_groups_0, pad = var_2262_pad_0, pad_type = var_2262_pad_type_0, strides = var_2262_strides_0, weight = layers_11_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_89_cast_fp16)[name = tensor<string, []>("op_2262_cast_fp16")]; |
| 1573 | tensor<fp16, [1, 768, 1, 1500]> obj_cast_fp16 = add(x = var_2256_cast_fp16, y = var_2262_cast_fp16)[name = tensor<string, []>("obj_cast_fp16")]; |
| 1574 | tensor<fp16, [1, 768, 1, 1500]> inputs_47_cast_fp16 = add(x = inputs_45_cast_fp16, y = obj_cast_fp16)[name = tensor<string, []>("inputs_47_cast_fp16")]; |
| 1575 | tensor<int32, [1]> out_47_axes_0 = const()[name = tensor<string, []>("out_47_axes_0"), val = tensor<int32, [1]>([1])]; |
| 1576 | tensor<fp16, []> var_2273_to_fp16 = const()[name = tensor<string, []>("op_2273_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 1577 | tensor<fp16, [1, 768, 1, 1500]> out_47_cast_fp16 = layer_norm(axes = out_47_axes_0, epsilon = var_2273_to_fp16, x = inputs_47_cast_fp16)[name = tensor<string, []>("out_47_cast_fp16")]; |
| 1578 | tensor<fp16, [768]> input_91_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_91_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59021312)))]; |
| 1579 | tensor<fp16, [768]> input_91_beta_0_to_fp16 = const()[name = tensor<string, []>("input_91_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59022912)))]; |
| 1580 | tensor<fp16, []> input_91_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_91_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 1581 | tensor<fp16, [1, 768, 1, 1500]> input_91_cast_fp16 = batch_norm(beta = input_91_beta_0_to_fp16, epsilon = input_91_epsilon_0_to_fp16, gamma = input_91_gamma_0_to_fp16, mean = var_57_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_47_cast_fp16)[name = tensor<string, []>("input_91_cast_fp16")]; |
| 1582 | tensor<string, []> var_2291_pad_type_0 = const()[name = tensor<string, []>("op_2291_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1583 | tensor<int32, [2]> var_2291_strides_0 = const()[name = tensor<string, []>("op_2291_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1584 | tensor<int32, [4]> var_2291_pad_0 = const()[name = tensor<string, []>("op_2291_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1585 | tensor<int32, [2]> var_2291_dilations_0 = const()[name = tensor<string, []>("op_2291_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1586 | tensor<int32, []> var_2291_groups_0 = const()[name = tensor<string, []>("op_2291_groups_0"), val = tensor<int32, []>(1)]; |
| 1587 | tensor<fp16, [3072, 768, 1, 1]> layers_11_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1179648]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59024512))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60204224))), name = tensor<string, []>("layers_11_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([3072, 768, 1, 1])]; |
| 1588 | tensor<fp16, [3072]> layers_11_fc1_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_11_fc1_inlier_module_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60204352)))]; |
| 1589 | tensor<fp16, [1, 3072, 1, 1500]> var_2291_cast_fp16 = conv(bias = layers_11_fc1_inlier_module_bias_to_fp16, dilations = var_2291_dilations_0, groups = var_2291_groups_0, pad = var_2291_pad_0, pad_type = var_2291_pad_type_0, strides = var_2291_strides_0, weight = layers_11_fc1_inlier_module_weight_to_fp16_palettized, x = input_91_cast_fp16)[name = tensor<string, []>("op_2291_cast_fp16")]; |
| 1590 | tensor<string, []> var_2297_pad_type_0 = const()[name = tensor<string, []>("op_2297_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1591 | tensor<int32, [2]> var_2297_strides_0 = const()[name = tensor<string, []>("op_2297_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1592 | tensor<int32, [4]> var_2297_pad_0 = const()[name = tensor<string, []>("op_2297_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1593 | tensor<int32, [2]> var_2297_dilations_0 = const()[name = tensor<string, []>("op_2297_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1594 | tensor<int32, []> var_2297_groups_0 = const()[name = tensor<string, []>("op_2297_groups_0"), val = tensor<int32, []>(1)]; |
| 1595 | tensor<fp16, [3072, 768, 1, 1]> layers_11_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60241728))), name = tensor<string, []>("layers_11_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [15540]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60210560))), shape = tensor<uint32, [4]>([3072, 768, 1, 1])]; |
| 1596 | tensor<fp16, [1, 3072, 1, 1500]> var_2297_cast_fp16 = conv(dilations = var_2297_dilations_0, groups = var_2297_groups_0, pad = var_2297_pad_0, pad_type = var_2297_pad_type_0, strides = var_2297_strides_0, weight = layers_11_fc1_outlier_module_weight_to_fp16_sparsified, x = input_91_cast_fp16)[name = tensor<string, []>("op_2297_cast_fp16")]; |
| 1597 | tensor<fp16, [1, 3072, 1, 1500]> input_93_cast_fp16 = add(x = var_2291_cast_fp16, y = var_2297_cast_fp16)[name = tensor<string, []>("input_93_cast_fp16")]; |
| 1598 | tensor<string, []> input_mode_0 = const()[name = tensor<string, []>("input_mode_0"), val = tensor<string, []>("EXACT")]; |
| 1599 | tensor<fp16, [1, 3072, 1, 1500]> input_cast_fp16 = gelu(mode = input_mode_0, x = input_93_cast_fp16)[name = tensor<string, []>("input_cast_fp16")]; |
| 1600 | tensor<string, []> var_2308_pad_type_0 = const()[name = tensor<string, []>("op_2308_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1601 | tensor<int32, [2]> var_2308_strides_0 = const()[name = tensor<string, []>("op_2308_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1602 | tensor<int32, [4]> var_2308_pad_0 = const()[name = tensor<string, []>("op_2308_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1603 | tensor<int32, [2]> var_2308_dilations_0 = const()[name = tensor<string, []>("op_2308_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1604 | tensor<int32, []> var_2308_groups_0 = const()[name = tensor<string, []>("op_2308_groups_0"), val = tensor<int32, []>(1)]; |
| 1605 | tensor<fp16, [768, 3072, 1, 1]> layers_11_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor<uint8, [1179648]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60536704))), lut = tensor<fp16, [16]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61716416))), name = tensor<string, []>("layers_11_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor<uint32, [4]>([768, 3072, 1, 1])]; |
| 1606 | tensor<fp16, [768]> layers_11_fc2_inlier_module_bias_to_fp16 = const()[name = tensor<string, []>("layers_11_fc2_inlier_module_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61716544)))]; |
| 1607 | tensor<fp16, [1, 768, 1, 1500]> var_2308_cast_fp16 = conv(bias = layers_11_fc2_inlier_module_bias_to_fp16, dilations = var_2308_dilations_0, groups = var_2308_groups_0, pad = var_2308_pad_0, pad_type = var_2308_pad_type_0, strides = var_2308_strides_0, weight = layers_11_fc2_inlier_module_weight_to_fp16_palettized, x = input_cast_fp16)[name = tensor<string, []>("op_2308_cast_fp16")]; |
| 1608 | tensor<string, []> var_2314_pad_type_0 = const()[name = tensor<string, []>("op_2314_pad_type_0"), val = tensor<string, []>("valid")]; |
| 1609 | tensor<int32, [2]> var_2314_strides_0 = const()[name = tensor<string, []>("op_2314_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1610 | tensor<int32, [4]> var_2314_pad_0 = const()[name = tensor<string, []>("op_2314_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 1611 | tensor<int32, [2]> var_2314_dilations_0 = const()[name = tensor<string, []>("op_2314_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| 1612 | tensor<int32, []> var_2314_groups_0 = const()[name = tensor<string, []>("op_2314_groups_0"), val = tensor<int32, []>(1)]; |
| 1613 | tensor<fp16, [768, 3072, 1, 1]> layers_11_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor<uint8, [294912]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61753984))), name = tensor<string, []>("layers_11_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor<fp16, [17871]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61718144))), shape = tensor<uint32, [4]>([768, 3072, 1, 1])]; |
| 1614 | tensor<fp16, [1, 768, 1, 1500]> var_2314_cast_fp16 = conv(dilations = var_2314_dilations_0, groups = var_2314_groups_0, pad = var_2314_pad_0, pad_type = var_2314_pad_type_0, strides = var_2314_strides_0, weight = layers_11_fc2_outlier_module_weight_to_fp16_sparsified, x = input_cast_fp16)[name = tensor<string, []>("op_2314_cast_fp16")]; |
| 1615 | tensor<fp16, [1, 768, 1, 1500]> hidden_states_cast_fp16 = add(x = var_2308_cast_fp16, y = var_2314_cast_fp16)[name = tensor<string, []>("hidden_states_cast_fp16")]; |
| 1616 | tensor<fp16, [1, 768, 1, 1500]> inputs_cast_fp16 = add(x = inputs_47_cast_fp16, y = hidden_states_cast_fp16)[name = tensor<string, []>("inputs_cast_fp16")]; |
| 1617 | tensor<int32, [1]> out_axes_0 = const()[name = tensor<string, []>("out_axes_0"), val = tensor<int32, [1]>([1])]; |
| 1618 | tensor<fp16, []> var_2329_to_fp16 = const()[name = tensor<string, []>("op_2329_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 1619 | tensor<fp16, [1, 768, 1, 1500]> out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_2329_to_fp16, x = inputs_cast_fp16)[name = tensor<string, []>("out_cast_fp16")]; |
| 1620 | tensor<fp16, [768]> encoder_output_embeds_type_fp32_gamma_0_to_fp16 = const()[name = tensor<string, []>("encoder_output_embeds_type_fp32_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62048960)))]; |
| 1621 | tensor<fp16, [768]> encoder_output_embeds_type_fp32_beta_0_to_fp16 = const()[name = tensor<string, []>("encoder_output_embeds_type_fp32_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62050560)))]; |
| 1622 | tensor<fp16, []> encoder_output_embeds_type_fp32_epsilon_0_to_fp16 = const()[name = tensor<string, []>("encoder_output_embeds_type_fp32_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 1623 | tensor<fp16, [1, 768, 1, 1500]> encoder_output_embeds = batch_norm(beta = encoder_output_embeds_type_fp32_beta_0_to_fp16, epsilon = encoder_output_embeds_type_fp32_epsilon_0_to_fp16, gamma = encoder_output_embeds_type_fp32_gamma_0_to_fp16, mean = var_57_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_cast_fp16)[name = tensor<string, []>("encoder_output_embeds_type_fp32_cast_fp16")]; |
| 1624 | } -> (encoder_output_embeds); |
| 1625 | } |