distil-whisper_distil-large-v3_594MB/TextDecoder.mlmodelc/model.mil
| 1 | program(1.0) |
| 2 | [buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "5.33.5"}, {"coremlc-version", "1877.40.3"}, {"coremltools-component-torch", "2.2.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "7.1"}})] |
| 3 | { |
| 4 | func main<ios16>(tensor<int32, [1]> cache_length, tensor<fp16, [1, 448]> decoder_key_padding_mask, tensor<fp16, [1, 1280, 1, 1500]> encoder_output_embeds, tensor<int32, [1]> input_ids, tensor<fp16, [1, 2560, 1, 448]> key_cache, tensor<fp16, [1, 448]> kv_cache_update_mask, tensor<fp16, [1, 2560, 1, 448]> value_cache) { |
| 5 | tensor<int32, []> var_20_axis_0 = const()[name = tensor<string, []>("op_20_axis_0"), val = tensor<int32, []>(0)]; |
| 6 | tensor<int32, []> var_20_batch_dims_0 = const()[name = tensor<string, []>("op_20_batch_dims_0"), val = tensor<int32, []>(0)]; |
| 7 | tensor<fp16, [51866, 1280]> embed_tokens_weight_to_fp16 = const()[name = tensor<string, []>("embed_tokens_weight_to_fp16"), val = tensor<fp16, [51866, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))]; |
| 8 | tensor<fp16, [1, 1280]> var_20_cast_fp16 = gather(axis = var_20_axis_0, batch_dims = var_20_batch_dims_0, indices = input_ids, x = embed_tokens_weight_to_fp16)[name = tensor<string, []>("op_20_cast_fp16")]; |
| 9 | tensor<int32, []> var_24_axis_0 = const()[name = tensor<string, []>("op_24_axis_0"), val = tensor<int32, []>(0)]; |
| 10 | tensor<int32, []> var_24_batch_dims_0 = const()[name = tensor<string, []>("op_24_batch_dims_0"), val = tensor<int32, []>(0)]; |
| 11 | tensor<fp16, [448, 1280]> embed_positions_weight_to_fp16 = const()[name = tensor<string, []>("embed_positions_weight_to_fp16"), val = tensor<fp16, [448, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132777088)))]; |
| 12 | tensor<fp16, [1, 1280]> var_24_cast_fp16 = gather(axis = var_24_axis_0, batch_dims = var_24_batch_dims_0, indices = cache_length, x = embed_positions_weight_to_fp16)[name = tensor<string, []>("op_24_cast_fp16")]; |
| 13 | tensor<fp16, [1, 1280]> hidden_states_1_cast_fp16 = add(x = var_20_cast_fp16, y = var_24_cast_fp16)[name = tensor<string, []>("hidden_states_1_cast_fp16")]; |
| 14 | tensor<int32, [1]> var_38_axes_0 = const()[name = tensor<string, []>("op_38_axes_0"), val = tensor<int32, [1]>([2])]; |
| 15 | tensor<fp16, [1, 1280, 1]> var_38_cast_fp16 = expand_dims(axes = var_38_axes_0, x = hidden_states_1_cast_fp16)[name = tensor<string, []>("op_38_cast_fp16")]; |
| 16 | tensor<int32, [1]> inputs_1_axes_0 = const()[name = tensor<string, []>("inputs_1_axes_0"), val = tensor<int32, [1]>([3])]; |
| 17 | tensor<fp16, [1, 1280, 1, 1]> inputs_1_cast_fp16 = expand_dims(axes = inputs_1_axes_0, x = var_38_cast_fp16)[name = tensor<string, []>("inputs_1_cast_fp16")]; |
| 18 | tensor<int32, [2]> tile_0 = const()[name = tensor<string, []>("tile_0"), val = tensor<int32, [2]>([1280, 1280])]; |
| 19 | tensor<int32, []> var_43_axis_0 = const()[name = tensor<string, []>("op_43_axis_0"), val = tensor<int32, []>(1)]; |
| 20 | tensor<fp16, [1, 1280, 1, 448]> var_43_cast_fp16_0, tensor<fp16, [1, 1280, 1, 448]> var_43_cast_fp16_1 = split(axis = var_43_axis_0, split_sizes = tile_0, x = key_cache)[name = tensor<string, []>("op_43_cast_fp16")]; |
| 21 | tensor<int32, [2]> tile_1 = const()[name = tensor<string, []>("tile_1"), val = tensor<int32, [2]>([1280, 1280])]; |
| 22 | tensor<int32, []> var_48_axis_0 = const()[name = tensor<string, []>("op_48_axis_0"), val = tensor<int32, []>(1)]; |
| 23 | tensor<fp16, [1, 1280, 1, 448]> var_48_cast_fp16_0, tensor<fp16, [1, 1280, 1, 448]> var_48_cast_fp16_1 = split(axis = var_48_axis_0, split_sizes = tile_1, x = value_cache)[name = tensor<string, []>("op_48_cast_fp16")]; |
| 24 | tensor<int32, []> var_56 = const()[name = tensor<string, []>("op_56"), val = tensor<int32, []>(3)]; |
| 25 | tensor<int32, []> var_63 = const()[name = tensor<string, []>("op_63"), val = tensor<int32, []>(1)]; |
| 26 | tensor<bool, []> var_64 = const()[name = tensor<string, []>("op_64"), val = tensor<bool, []>(true)]; |
| 27 | tensor<int32, [1]> var_76 = const()[name = tensor<string, []>("op_76"), val = tensor<int32, [1]>([1])]; |
| 28 | tensor<fp16, [1, 1, 1, 1]> channels_mean_1_cast_fp16 = reduce_mean(axes = var_76, keep_dims = var_64, x = inputs_1_cast_fp16)[name = tensor<string, []>("channels_mean_1_cast_fp16")]; |
| 29 | tensor<fp16, [1, 1280, 1, 1]> zero_mean_1_cast_fp16 = sub(x = inputs_1_cast_fp16, y = channels_mean_1_cast_fp16)[name = tensor<string, []>("zero_mean_1_cast_fp16")]; |
| 30 | tensor<fp16, [1, 1280, 1, 1]> zero_mean_sq_1_cast_fp16 = mul(x = zero_mean_1_cast_fp16, y = zero_mean_1_cast_fp16)[name = tensor<string, []>("zero_mean_sq_1_cast_fp16")]; |
| 31 | tensor<int32, [1]> var_80 = const()[name = tensor<string, []>("op_80"), val = tensor<int32, [1]>([1])]; |
| 32 | tensor<fp16, [1, 1, 1, 1]> var_81_cast_fp16 = reduce_mean(axes = var_80, keep_dims = var_64, x = zero_mean_sq_1_cast_fp16)[name = tensor<string, []>("op_81_cast_fp16")]; |
| 33 | tensor<fp16, []> var_82_to_fp16 = const()[name = tensor<string, []>("op_82_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 34 | tensor<fp16, [1, 1, 1, 1]> var_83_cast_fp16 = add(x = var_81_cast_fp16, y = var_82_to_fp16)[name = tensor<string, []>("op_83_cast_fp16")]; |
| 35 | tensor<fp16, []> denom_1_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_1_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)]; |
| 36 | tensor<fp16, [1, 1, 1, 1]> denom_1_cast_fp16 = rsqrt(epsilon = denom_1_epsilon_0_to_fp16, x = var_83_cast_fp16)[name = tensor<string, []>("denom_1_cast_fp16")]; |
| 37 | tensor<fp16, [1, 1280, 1, 1]> out_1_cast_fp16 = mul(x = zero_mean_1_cast_fp16, y = denom_1_cast_fp16)[name = tensor<string, []>("out_1_cast_fp16")]; |
| 38 | tensor<fp16, [1280]> obj_1_mean_0_to_fp16 = const()[name = tensor<string, []>("obj_1_mean_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133924032)))]; |
| 39 | tensor<fp16, [1280]> obj_1_variance_0_to_fp16 = const()[name = tensor<string, []>("obj_1_variance_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133926656)))]; |
| 40 | tensor<fp16, [1280]> obj_1_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_1_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133929280)))]; |
| 41 | tensor<fp16, [1280]> obj_1_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_1_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133931904)))]; |
| 42 | tensor<fp16, []> obj_1_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_1_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 43 | tensor<fp16, [1, 1280, 1, 1]> 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 = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_1_cast_fp16)[name = tensor<string, []>("obj_1_cast_fp16")]; |
| 44 | tensor<int32, [2]> var_98 = const()[name = tensor<string, []>("op_98"), val = tensor<int32, [2]>([1, 1])]; |
| 45 | tensor<int32, [2]> var_100 = const()[name = tensor<string, []>("op_100"), val = tensor<int32, [2]>([1, 1])]; |
| 46 | tensor<string, []> query_1_pad_type_0 = const()[name = tensor<string, []>("query_1_pad_type_0"), val = tensor<string, []>("custom")]; |
| 47 | tensor<int32, [4]> query_1_pad_0 = const()[name = tensor<string, []>("query_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 48 | tensor<fp16, [1280, 1280, 1, 1]> layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133934528)))]; |
| 49 | tensor<fp16, [1280]> layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137211392)))]; |
| 50 | tensor<fp16, [1, 1280, 1, 1]> query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = var_100, groups = var_63, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = var_98, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = tensor<string, []>("query_1_cast_fp16")]; |
| 51 | tensor<int32, [2]> var_104 = const()[name = tensor<string, []>("op_104"), val = tensor<int32, [2]>([1, 1])]; |
| 52 | tensor<int32, [2]> var_106 = const()[name = tensor<string, []>("op_106"), val = tensor<int32, [2]>([1, 1])]; |
| 53 | tensor<string, []> current_key_1_pad_type_0 = const()[name = tensor<string, []>("current_key_1_pad_type_0"), val = tensor<string, []>("custom")]; |
| 54 | tensor<int32, [4]> current_key_1_pad_0 = const()[name = tensor<string, []>("current_key_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 55 | tensor<fp16, [1280, 1280, 1, 1]> layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137214016)))]; |
| 56 | tensor<fp16, [1, 1280, 1, 1]> current_key_1_cast_fp16 = conv(dilations = var_106, groups = var_63, pad = current_key_1_pad_0, pad_type = current_key_1_pad_type_0, strides = var_104, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = tensor<string, []>("current_key_1_cast_fp16")]; |
| 57 | tensor<int32, [2]> var_111 = const()[name = tensor<string, []>("op_111"), val = tensor<int32, [2]>([1, 1])]; |
| 58 | tensor<int32, [2]> var_113 = const()[name = tensor<string, []>("op_113"), val = tensor<int32, [2]>([1, 1])]; |
| 59 | tensor<string, []> current_value_1_pad_type_0 = const()[name = tensor<string, []>("current_value_1_pad_type_0"), val = tensor<string, []>("custom")]; |
| 60 | tensor<int32, [4]> current_value_1_pad_0 = const()[name = tensor<string, []>("current_value_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 61 | tensor<fp16, [1280, 1280, 1, 1]> layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(140490880)))]; |
| 62 | tensor<fp16, [1280]> layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(143767744)))]; |
| 63 | tensor<fp16, [1, 1280, 1, 1]> current_value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = var_113, groups = var_63, pad = current_value_1_pad_0, pad_type = current_value_1_pad_type_0, strides = var_111, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = tensor<string, []>("current_value_1_cast_fp16")]; |
| 64 | tensor<int32, [1]> var_117_axes_0 = const()[name = tensor<string, []>("op_117_axes_0"), val = tensor<int32, [1]>([1])]; |
| 65 | tensor<fp16, [1, 1, 448]> var_117_cast_fp16 = expand_dims(axes = var_117_axes_0, x = kv_cache_update_mask)[name = tensor<string, []>("op_117_cast_fp16")]; |
| 66 | tensor<int32, [1]> var_118_axes_0 = const()[name = tensor<string, []>("op_118_axes_0"), val = tensor<int32, [1]>([2])]; |
| 67 | tensor<fp16, [1, 1, 1, 448]> var_118_cast_fp16 = expand_dims(axes = var_118_axes_0, x = var_117_cast_fp16)[name = tensor<string, []>("op_118_cast_fp16")]; |
| 68 | tensor<fp16, [1, 1280, 1, 448]> var_120_cast_fp16 = mul(x = current_key_1_cast_fp16, y = var_118_cast_fp16)[name = tensor<string, []>("op_120_cast_fp16")]; |
| 69 | tensor<fp16, []> var_57_to_fp16 = const()[name = tensor<string, []>("op_57_to_fp16"), val = tensor<fp16, []>(0x1p+0)]; |
| 70 | tensor<fp16, [1, 1, 1, 448]> var_121_cast_fp16 = sub(x = var_57_to_fp16, y = var_118_cast_fp16)[name = tensor<string, []>("op_121_cast_fp16")]; |
| 71 | tensor<fp16, [1, 1280, 1, 448]> var_122_cast_fp16 = mul(x = var_43_cast_fp16_0, y = var_121_cast_fp16)[name = tensor<string, []>("op_122_cast_fp16")]; |
| 72 | tensor<fp16, [1, 1280, 1, 448]> key_1_cast_fp16 = add(x = var_120_cast_fp16, y = var_122_cast_fp16)[name = tensor<string, []>("key_1_cast_fp16")]; |
| 73 | tensor<fp16, [1, 1280, 1, 448]> var_124_cast_fp16 = mul(x = current_value_1_cast_fp16, y = var_118_cast_fp16)[name = tensor<string, []>("op_124_cast_fp16")]; |
| 74 | tensor<fp16, [1, 1280, 1, 448]> var_126_cast_fp16 = mul(x = var_48_cast_fp16_0, y = var_121_cast_fp16)[name = tensor<string, []>("op_126_cast_fp16")]; |
| 75 | tensor<fp16, [1, 1280, 1, 448]> value_1_cast_fp16 = add(x = var_124_cast_fp16, y = var_126_cast_fp16)[name = tensor<string, []>("value_1_cast_fp16")]; |
| 76 | tensor<int32, [4]> var_129 = const()[name = tensor<string, []>("op_129"), val = tensor<int32, [4]>([1, 20, 64, -1])]; |
| 77 | tensor<fp16, [1, 20, 64, 1]> var_130_cast_fp16 = reshape(shape = var_129, x = query_1_cast_fp16)[name = tensor<string, []>("op_130_cast_fp16")]; |
| 78 | tensor<fp16, []> var_131_to_fp16 = const()[name = tensor<string, []>("op_131_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| 79 | tensor<fp16, [1, 20, 64, 1]> var_132_cast_fp16 = mul(x = var_130_cast_fp16, y = var_131_to_fp16)[name = tensor<string, []>("op_132_cast_fp16")]; |
| 80 | tensor<int32, [4]> var_133 = const()[name = tensor<string, []>("op_133"), val = tensor<int32, [4]>([1, 20, 64, -1])]; |
| 81 | tensor<fp16, [1, 20, 64, 448]> var_134_cast_fp16 = reshape(shape = var_133, x = key_1_cast_fp16)[name = tensor<string, []>("op_134_cast_fp16")]; |
| 82 | tensor<bool, []> mh_w_1_transpose_x_0 = const()[name = tensor<string, []>("mh_w_1_transpose_x_0"), val = tensor<bool, []>(true)]; |
| 83 | tensor<bool, []> mh_w_1_transpose_y_0 = const()[name = tensor<string, []>("mh_w_1_transpose_y_0"), val = tensor<bool, []>(false)]; |
| 84 | tensor<fp16, [1, 20, 1, 448]> mh_w_1_cast_fp16 = matmul(transpose_x = mh_w_1_transpose_x_0, transpose_y = mh_w_1_transpose_y_0, x = var_132_cast_fp16, y = var_134_cast_fp16)[name = tensor<string, []>("mh_w_1_cast_fp16")]; |
| 85 | tensor<int32, [1]> var_138_axes_0 = const()[name = tensor<string, []>("op_138_axes_0"), val = tensor<int32, [1]>([1])]; |
| 86 | tensor<fp16, [1, 1, 448]> var_138_cast_fp16 = expand_dims(axes = var_138_axes_0, x = decoder_key_padding_mask)[name = tensor<string, []>("op_138_cast_fp16")]; |
| 87 | tensor<int32, [1]> var_139_axes_0 = const()[name = tensor<string, []>("op_139_axes_0"), val = tensor<int32, [1]>([2])]; |
| 88 | tensor<fp16, [1, 1, 1, 448]> var_139_cast_fp16 = expand_dims(axes = var_139_axes_0, x = var_138_cast_fp16)[name = tensor<string, []>("op_139_cast_fp16")]; |
| 89 | tensor<fp16, [1, 20, 1, 448]> mh_w_3_cast_fp16 = add(x = mh_w_1_cast_fp16, y = var_139_cast_fp16)[name = tensor<string, []>("mh_w_3_cast_fp16")]; |
| 90 | tensor<fp16, [1, 20, 1, 448]> var_142_cast_fp16 = softmax(axis = var_56, x = mh_w_3_cast_fp16)[name = tensor<string, []>("op_142_cast_fp16")]; |
| 91 | tensor<int32, [4]> var_143 = const()[name = tensor<string, []>("op_143"), val = tensor<int32, [4]>([1, 20, 64, -1])]; |
| 92 | tensor<fp16, [1, 20, 64, 448]> var_144_cast_fp16 = reshape(shape = var_143, x = value_1_cast_fp16)[name = tensor<string, []>("op_144_cast_fp16")]; |
| 93 | tensor<bool, []> attn_1_transpose_x_0 = const()[name = tensor<string, []>("attn_1_transpose_x_0"), val = tensor<bool, []>(false)]; |
| 94 | tensor<bool, []> attn_1_transpose_y_0 = const()[name = tensor<string, []>("attn_1_transpose_y_0"), val = tensor<bool, []>(true)]; |
| 95 | tensor<fp16, [1, 20, 64, 1]> attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_144_cast_fp16, y = var_142_cast_fp16)[name = tensor<string, []>("attn_1_cast_fp16")]; |
| 96 | tensor<int32, [4]> var_147 = const()[name = tensor<string, []>("op_147"), val = tensor<int32, [4]>([1, 1280, 1, -1])]; |
| 97 | tensor<fp16, [1, 1280, 1, 1]> input_1_cast_fp16 = reshape(shape = var_147, x = attn_1_cast_fp16)[name = tensor<string, []>("input_1_cast_fp16")]; |
| 98 | tensor<int32, [2]> var_151 = const()[name = tensor<string, []>("op_151"), val = tensor<int32, [2]>([1, 1])]; |
| 99 | tensor<int32, [2]> var_153 = const()[name = tensor<string, []>("op_153"), val = tensor<int32, [2]>([1, 1])]; |
| 100 | tensor<string, []> obj_7_pad_type_0 = const()[name = tensor<string, []>("obj_7_pad_type_0"), val = tensor<string, []>("custom")]; |
| 101 | tensor<int32, [4]> obj_7_pad_0 = const()[name = tensor<string, []>("obj_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 102 | tensor<fp16, [1280, 1280, 1, 1]> layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(143770368)))]; |
| 103 | tensor<fp16, [1280]> layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147047232)))]; |
| 104 | tensor<fp16, [1, 1280, 1, 1]> obj_7_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = var_153, groups = var_63, pad = obj_7_pad_0, pad_type = obj_7_pad_type_0, strides = var_151, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = tensor<string, []>("obj_7_cast_fp16")]; |
| 105 | tensor<fp16, [1, 1280, 1, 1]> inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_7_cast_fp16)[name = tensor<string, []>("inputs_3_cast_fp16")]; |
| 106 | tensor<int32, [1]> var_163 = const()[name = tensor<string, []>("op_163"), val = tensor<int32, [1]>([1])]; |
| 107 | tensor<fp16, [1, 1, 1, 1]> channels_mean_3_cast_fp16 = reduce_mean(axes = var_163, keep_dims = var_64, x = inputs_3_cast_fp16)[name = tensor<string, []>("channels_mean_3_cast_fp16")]; |
| 108 | tensor<fp16, [1, 1280, 1, 1]> zero_mean_3_cast_fp16 = sub(x = inputs_3_cast_fp16, y = channels_mean_3_cast_fp16)[name = tensor<string, []>("zero_mean_3_cast_fp16")]; |
| 109 | tensor<fp16, [1, 1280, 1, 1]> zero_mean_sq_3_cast_fp16 = mul(x = zero_mean_3_cast_fp16, y = zero_mean_3_cast_fp16)[name = tensor<string, []>("zero_mean_sq_3_cast_fp16")]; |
| 110 | tensor<int32, [1]> var_167 = const()[name = tensor<string, []>("op_167"), val = tensor<int32, [1]>([1])]; |
| 111 | tensor<fp16, [1, 1, 1, 1]> var_168_cast_fp16 = reduce_mean(axes = var_167, keep_dims = var_64, x = zero_mean_sq_3_cast_fp16)[name = tensor<string, []>("op_168_cast_fp16")]; |
| 112 | tensor<fp16, []> var_169_to_fp16 = const()[name = tensor<string, []>("op_169_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 113 | tensor<fp16, [1, 1, 1, 1]> var_170_cast_fp16 = add(x = var_168_cast_fp16, y = var_169_to_fp16)[name = tensor<string, []>("op_170_cast_fp16")]; |
| 114 | tensor<fp16, []> denom_3_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_3_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)]; |
| 115 | tensor<fp16, [1, 1, 1, 1]> denom_3_cast_fp16 = rsqrt(epsilon = denom_3_epsilon_0_to_fp16, x = var_170_cast_fp16)[name = tensor<string, []>("denom_3_cast_fp16")]; |
| 116 | tensor<fp16, [1, 1280, 1, 1]> out_3_cast_fp16 = mul(x = zero_mean_3_cast_fp16, y = denom_3_cast_fp16)[name = tensor<string, []>("out_3_cast_fp16")]; |
| 117 | tensor<fp16, [1280]> obj_9_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_9_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147049856)))]; |
| 118 | tensor<fp16, [1280]> obj_9_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_9_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147052480)))]; |
| 119 | tensor<fp16, []> obj_9_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_9_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 120 | tensor<fp16, [1, 1280, 1, 1]> 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 = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_3_cast_fp16)[name = tensor<string, []>("obj_9_cast_fp16")]; |
| 121 | tensor<int32, [2]> var_185 = const()[name = tensor<string, []>("op_185"), val = tensor<int32, [2]>([1, 1])]; |
| 122 | tensor<int32, [2]> var_187 = const()[name = tensor<string, []>("op_187"), val = tensor<int32, [2]>([1, 1])]; |
| 123 | tensor<string, []> query_3_pad_type_0 = const()[name = tensor<string, []>("query_3_pad_type_0"), val = tensor<string, []>("custom")]; |
| 124 | tensor<int32, [4]> query_3_pad_0 = const()[name = tensor<string, []>("query_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 125 | tensor<fp16, [1280, 1280, 1, 1]> layers_0_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_encoder_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147055104)))]; |
| 126 | tensor<fp16, [1280]> layers_0_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150331968)))]; |
| 127 | tensor<fp16, [1, 1280, 1, 1]> query_3_cast_fp16 = conv(bias = layers_0_encoder_attn_q_proj_bias_to_fp16, dilations = var_187, groups = var_63, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = var_185, weight = layers_0_encoder_attn_q_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = tensor<string, []>("query_3_cast_fp16")]; |
| 128 | tensor<int32, [2]> var_191 = const()[name = tensor<string, []>("op_191"), val = tensor<int32, [2]>([1, 1])]; |
| 129 | tensor<int32, [2]> var_193 = const()[name = tensor<string, []>("op_193"), val = tensor<int32, [2]>([1, 1])]; |
| 130 | tensor<string, []> key_3_pad_type_0 = const()[name = tensor<string, []>("key_3_pad_type_0"), val = tensor<string, []>("custom")]; |
| 131 | tensor<int32, [4]> key_3_pad_0 = const()[name = tensor<string, []>("key_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 132 | tensor<fp16, [1280, 1280, 1, 1]> layers_0_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_encoder_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150334592)))]; |
| 133 | tensor<fp16, [1, 1280, 1, 1500]> key_3_cast_fp16 = conv(dilations = var_193, groups = var_63, pad = key_3_pad_0, pad_type = key_3_pad_type_0, strides = var_191, weight = layers_0_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("key_3_cast_fp16")]; |
| 134 | tensor<int32, [2]> var_198 = const()[name = tensor<string, []>("op_198"), val = tensor<int32, [2]>([1, 1])]; |
| 135 | tensor<int32, [2]> var_200 = const()[name = tensor<string, []>("op_200"), val = tensor<int32, [2]>([1, 1])]; |
| 136 | tensor<string, []> value_3_pad_type_0 = const()[name = tensor<string, []>("value_3_pad_type_0"), val = tensor<string, []>("custom")]; |
| 137 | tensor<int32, [4]> value_3_pad_0 = const()[name = tensor<string, []>("value_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 138 | tensor<fp16, [1280, 1280, 1, 1]> layers_0_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_encoder_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153611456)))]; |
| 139 | tensor<fp16, [1280]> layers_0_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(156888320)))]; |
| 140 | tensor<fp16, [1, 1280, 1, 1500]> value_3_cast_fp16 = conv(bias = layers_0_encoder_attn_v_proj_bias_to_fp16, dilations = var_200, groups = var_63, pad = value_3_pad_0, pad_type = value_3_pad_type_0, strides = var_198, weight = layers_0_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("value_3_cast_fp16")]; |
| 141 | tensor<int32, [4]> var_204 = const()[name = tensor<string, []>("op_204"), val = tensor<int32, [4]>([1, 20, 64, -1])]; |
| 142 | tensor<fp16, [1, 20, 64, 1]> var_205_cast_fp16 = reshape(shape = var_204, x = query_3_cast_fp16)[name = tensor<string, []>("op_205_cast_fp16")]; |
| 143 | tensor<fp16, []> var_206_to_fp16 = const()[name = tensor<string, []>("op_206_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| 144 | tensor<fp16, [1, 20, 64, 1]> var_207_cast_fp16 = mul(x = var_205_cast_fp16, y = var_206_to_fp16)[name = tensor<string, []>("op_207_cast_fp16")]; |
| 145 | tensor<int32, [4]> var_208 = const()[name = tensor<string, []>("op_208"), val = tensor<int32, [4]>([1, 20, 64, -1])]; |
| 146 | tensor<fp16, [1, 20, 64, 1500]> var_209_cast_fp16 = reshape(shape = var_208, x = key_3_cast_fp16)[name = tensor<string, []>("op_209_cast_fp16")]; |
| 147 | tensor<bool, []> mh_w_5_transpose_x_0 = const()[name = tensor<string, []>("mh_w_5_transpose_x_0"), val = tensor<bool, []>(true)]; |
| 148 | tensor<bool, []> mh_w_5_transpose_y_0 = const()[name = tensor<string, []>("mh_w_5_transpose_y_0"), val = tensor<bool, []>(false)]; |
| 149 | tensor<fp16, [1, 20, 1, 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_207_cast_fp16, y = var_209_cast_fp16)[name = tensor<string, []>("mh_w_5_cast_fp16")]; |
| 150 | tensor<fp16, [1, 20, 1, 1500]> obj_13_cast_fp16 = softmax(axis = var_56, x = mh_w_5_cast_fp16)[name = tensor<string, []>("obj_13_cast_fp16")]; |
| 151 | tensor<int32, [4]> var_213 = const()[name = tensor<string, []>("op_213"), val = tensor<int32, [4]>([1, 20, 64, -1])]; |
| 152 | tensor<fp16, [1, 20, 64, 1500]> var_214_cast_fp16 = reshape(shape = var_213, x = value_3_cast_fp16)[name = tensor<string, []>("op_214_cast_fp16")]; |
| 153 | tensor<bool, []> attn_3_transpose_x_0 = const()[name = tensor<string, []>("attn_3_transpose_x_0"), val = tensor<bool, []>(false)]; |
| 154 | tensor<bool, []> attn_3_transpose_y_0 = const()[name = tensor<string, []>("attn_3_transpose_y_0"), val = tensor<bool, []>(true)]; |
| 155 | tensor<fp16, [1, 20, 64, 1]> attn_3_cast_fp16 = matmul(transpose_x = attn_3_transpose_x_0, transpose_y = attn_3_transpose_y_0, x = var_214_cast_fp16, y = obj_13_cast_fp16)[name = tensor<string, []>("attn_3_cast_fp16")]; |
| 156 | tensor<int32, [4]> var_217 = const()[name = tensor<string, []>("op_217"), val = tensor<int32, [4]>([1, 1280, 1, -1])]; |
| 157 | tensor<fp16, [1, 1280, 1, 1]> input_3_cast_fp16 = reshape(shape = var_217, x = attn_3_cast_fp16)[name = tensor<string, []>("input_3_cast_fp16")]; |
| 158 | tensor<int32, [2]> var_221 = const()[name = tensor<string, []>("op_221"), val = tensor<int32, [2]>([1, 1])]; |
| 159 | tensor<int32, [2]> var_223 = const()[name = tensor<string, []>("op_223"), val = tensor<int32, [2]>([1, 1])]; |
| 160 | tensor<string, []> obj_11_pad_type_0 = const()[name = tensor<string, []>("obj_11_pad_type_0"), val = tensor<string, []>("custom")]; |
| 161 | tensor<int32, [4]> obj_11_pad_0 = const()[name = tensor<string, []>("obj_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 162 | tensor<fp16, [1280, 1280, 1, 1]> layers_0_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_encoder_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(156890944)))]; |
| 163 | tensor<fp16, [1280]> layers_0_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160167808)))]; |
| 164 | tensor<fp16, [1, 1280, 1, 1]> obj_11_cast_fp16 = conv(bias = layers_0_encoder_attn_o_proj_bias_to_fp16, dilations = var_223, groups = var_63, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = var_221, weight = layers_0_encoder_attn_o_proj_weight_to_fp16, x = input_3_cast_fp16)[name = tensor<string, []>("obj_11_cast_fp16")]; |
| 165 | tensor<fp16, [1, 1280, 1, 1]> inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = obj_11_cast_fp16)[name = tensor<string, []>("inputs_5_cast_fp16")]; |
| 166 | tensor<int32, [1]> var_229 = const()[name = tensor<string, []>("op_229"), val = tensor<int32, [1]>([1])]; |
| 167 | tensor<fp16, [1, 1, 1, 1]> channels_mean_5_cast_fp16 = reduce_mean(axes = var_229, keep_dims = var_64, x = inputs_5_cast_fp16)[name = tensor<string, []>("channels_mean_5_cast_fp16")]; |
| 168 | tensor<fp16, [1, 1280, 1, 1]> zero_mean_5_cast_fp16 = sub(x = inputs_5_cast_fp16, y = channels_mean_5_cast_fp16)[name = tensor<string, []>("zero_mean_5_cast_fp16")]; |
| 169 | tensor<fp16, [1, 1280, 1, 1]> zero_mean_sq_5_cast_fp16 = mul(x = zero_mean_5_cast_fp16, y = zero_mean_5_cast_fp16)[name = tensor<string, []>("zero_mean_sq_5_cast_fp16")]; |
| 170 | tensor<int32, [1]> var_233 = const()[name = tensor<string, []>("op_233"), val = tensor<int32, [1]>([1])]; |
| 171 | tensor<fp16, [1, 1, 1, 1]> var_234_cast_fp16 = reduce_mean(axes = var_233, keep_dims = var_64, x = zero_mean_sq_5_cast_fp16)[name = tensor<string, []>("op_234_cast_fp16")]; |
| 172 | tensor<fp16, []> var_235_to_fp16 = const()[name = tensor<string, []>("op_235_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 173 | tensor<fp16, [1, 1, 1, 1]> var_236_cast_fp16 = add(x = var_234_cast_fp16, y = var_235_to_fp16)[name = tensor<string, []>("op_236_cast_fp16")]; |
| 174 | tensor<fp16, []> denom_5_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_5_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)]; |
| 175 | tensor<fp16, [1, 1, 1, 1]> denom_5_cast_fp16 = rsqrt(epsilon = denom_5_epsilon_0_to_fp16, x = var_236_cast_fp16)[name = tensor<string, []>("denom_5_cast_fp16")]; |
| 176 | tensor<fp16, [1, 1280, 1, 1]> out_5_cast_fp16 = mul(x = zero_mean_5_cast_fp16, y = denom_5_cast_fp16)[name = tensor<string, []>("out_5_cast_fp16")]; |
| 177 | tensor<fp16, [1280]> input_5_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_5_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160170432)))]; |
| 178 | tensor<fp16, [1280]> input_5_beta_0_to_fp16 = const()[name = tensor<string, []>("input_5_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160173056)))]; |
| 179 | tensor<fp16, []> input_5_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_5_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 180 | tensor<fp16, [1, 1280, 1, 1]> input_5_cast_fp16 = batch_norm(beta = input_5_beta_0_to_fp16, epsilon = input_5_epsilon_0_to_fp16, gamma = input_5_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_5_cast_fp16)[name = tensor<string, []>("input_5_cast_fp16")]; |
| 181 | tensor<int32, [2]> var_247 = const()[name = tensor<string, []>("op_247"), val = tensor<int32, [2]>([1, 1])]; |
| 182 | tensor<int32, [2]> var_249 = const()[name = tensor<string, []>("op_249"), val = tensor<int32, [2]>([1, 1])]; |
| 183 | tensor<string, []> input_7_pad_type_0 = const()[name = tensor<string, []>("input_7_pad_type_0"), val = tensor<string, []>("custom")]; |
| 184 | tensor<int32, [4]> input_7_pad_0 = const()[name = tensor<string, []>("input_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 185 | tensor<fp16, [5120, 1280, 1, 1]> layers_0_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160175680)))]; |
| 186 | tensor<fp16, [5120]> layers_0_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(173282944)))]; |
| 187 | tensor<fp16, [1, 5120, 1, 1]> input_7_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = var_249, groups = var_63, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = var_247, weight = layers_0_fc1_weight_to_fp16, x = input_5_cast_fp16)[name = tensor<string, []>("input_7_cast_fp16")]; |
| 188 | tensor<string, []> input_9_mode_0 = const()[name = tensor<string, []>("input_9_mode_0"), val = tensor<string, []>("EXACT")]; |
| 189 | tensor<fp16, [1, 5120, 1, 1]> input_9_cast_fp16 = gelu(mode = input_9_mode_0, x = input_7_cast_fp16)[name = tensor<string, []>("input_9_cast_fp16")]; |
| 190 | tensor<int32, [2]> var_255 = const()[name = tensor<string, []>("op_255"), val = tensor<int32, [2]>([1, 1])]; |
| 191 | tensor<int32, [2]> var_257 = const()[name = tensor<string, []>("op_257"), val = tensor<int32, [2]>([1, 1])]; |
| 192 | tensor<string, []> hidden_states_3_pad_type_0 = const()[name = tensor<string, []>("hidden_states_3_pad_type_0"), val = tensor<string, []>("custom")]; |
| 193 | tensor<int32, [4]> hidden_states_3_pad_0 = const()[name = tensor<string, []>("hidden_states_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 194 | tensor<fp16, [1280, 5120, 1, 1]> layers_0_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_0_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(173293248)))]; |
| 195 | tensor<fp16, [1280]> layers_0_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_0_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186400512)))]; |
| 196 | tensor<fp16, [1, 1280, 1, 1]> hidden_states_3_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = var_257, groups = var_63, pad = hidden_states_3_pad_0, pad_type = hidden_states_3_pad_type_0, strides = var_255, weight = layers_0_fc2_weight_to_fp16, x = input_9_cast_fp16)[name = tensor<string, []>("hidden_states_3_cast_fp16")]; |
| 197 | tensor<fp16, [1, 1280, 1, 1]> inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = hidden_states_3_cast_fp16)[name = tensor<string, []>("inputs_7_cast_fp16")]; |
| 198 | tensor<int32, []> var_270 = const()[name = tensor<string, []>("op_270"), val = tensor<int32, []>(3)]; |
| 199 | tensor<int32, []> var_277 = const()[name = tensor<string, []>("op_277"), val = tensor<int32, []>(1)]; |
| 200 | tensor<bool, []> var_278 = const()[name = tensor<string, []>("op_278"), val = tensor<bool, []>(true)]; |
| 201 | tensor<int32, [1]> var_290 = const()[name = tensor<string, []>("op_290"), val = tensor<int32, [1]>([1])]; |
| 202 | tensor<fp16, [1, 1, 1, 1]> channels_mean_7_cast_fp16 = reduce_mean(axes = var_290, keep_dims = var_278, x = inputs_7_cast_fp16)[name = tensor<string, []>("channels_mean_7_cast_fp16")]; |
| 203 | tensor<fp16, [1, 1280, 1, 1]> zero_mean_7_cast_fp16 = sub(x = inputs_7_cast_fp16, y = channels_mean_7_cast_fp16)[name = tensor<string, []>("zero_mean_7_cast_fp16")]; |
| 204 | tensor<fp16, [1, 1280, 1, 1]> zero_mean_sq_7_cast_fp16 = mul(x = zero_mean_7_cast_fp16, y = zero_mean_7_cast_fp16)[name = tensor<string, []>("zero_mean_sq_7_cast_fp16")]; |
| 205 | tensor<int32, [1]> var_294 = const()[name = tensor<string, []>("op_294"), val = tensor<int32, [1]>([1])]; |
| 206 | tensor<fp16, [1, 1, 1, 1]> var_295_cast_fp16 = reduce_mean(axes = var_294, keep_dims = var_278, x = zero_mean_sq_7_cast_fp16)[name = tensor<string, []>("op_295_cast_fp16")]; |
| 207 | tensor<fp16, []> var_296_to_fp16 = const()[name = tensor<string, []>("op_296_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 208 | tensor<fp16, [1, 1, 1, 1]> var_297_cast_fp16 = add(x = var_295_cast_fp16, y = var_296_to_fp16)[name = tensor<string, []>("op_297_cast_fp16")]; |
| 209 | tensor<fp16, []> denom_7_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_7_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)]; |
| 210 | tensor<fp16, [1, 1, 1, 1]> denom_7_cast_fp16 = rsqrt(epsilon = denom_7_epsilon_0_to_fp16, x = var_297_cast_fp16)[name = tensor<string, []>("denom_7_cast_fp16")]; |
| 211 | tensor<fp16, [1, 1280, 1, 1]> out_7_cast_fp16 = mul(x = zero_mean_7_cast_fp16, y = denom_7_cast_fp16)[name = tensor<string, []>("out_7_cast_fp16")]; |
| 212 | tensor<fp16, [1280]> obj_15_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_15_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186403136)))]; |
| 213 | tensor<fp16, [1280]> obj_15_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_15_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186405760)))]; |
| 214 | tensor<fp16, []> obj_15_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_15_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 215 | tensor<fp16, [1, 1280, 1, 1]> obj_15_cast_fp16 = batch_norm(beta = obj_15_beta_0_to_fp16, epsilon = obj_15_epsilon_0_to_fp16, gamma = obj_15_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_7_cast_fp16)[name = tensor<string, []>("obj_15_cast_fp16")]; |
| 216 | tensor<int32, [2]> var_312 = const()[name = tensor<string, []>("op_312"), val = tensor<int32, [2]>([1, 1])]; |
| 217 | tensor<int32, [2]> var_314 = const()[name = tensor<string, []>("op_314"), val = tensor<int32, [2]>([1, 1])]; |
| 218 | tensor<string, []> query_5_pad_type_0 = const()[name = tensor<string, []>("query_5_pad_type_0"), val = tensor<string, []>("custom")]; |
| 219 | tensor<int32, [4]> query_5_pad_0 = const()[name = tensor<string, []>("query_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 220 | tensor<fp16, [1280, 1280, 1, 1]> layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186408384)))]; |
| 221 | tensor<fp16, [1280]> layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(189685248)))]; |
| 222 | tensor<fp16, [1, 1280, 1, 1]> query_5_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = var_314, groups = var_277, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = var_312, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_15_cast_fp16)[name = tensor<string, []>("query_5_cast_fp16")]; |
| 223 | tensor<int32, [2]> var_318 = const()[name = tensor<string, []>("op_318"), val = tensor<int32, [2]>([1, 1])]; |
| 224 | tensor<int32, [2]> var_320 = const()[name = tensor<string, []>("op_320"), val = tensor<int32, [2]>([1, 1])]; |
| 225 | tensor<string, []> current_key_pad_type_0 = const()[name = tensor<string, []>("current_key_pad_type_0"), val = tensor<string, []>("custom")]; |
| 226 | tensor<int32, [4]> current_key_pad_0 = const()[name = tensor<string, []>("current_key_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 227 | tensor<fp16, [1280, 1280, 1, 1]> layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(189687872)))]; |
| 228 | tensor<fp16, [1, 1280, 1, 1]> current_key_cast_fp16 = conv(dilations = var_320, groups = var_277, pad = current_key_pad_0, pad_type = current_key_pad_type_0, strides = var_318, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_15_cast_fp16)[name = tensor<string, []>("current_key_cast_fp16")]; |
| 229 | tensor<int32, [2]> var_325 = const()[name = tensor<string, []>("op_325"), val = tensor<int32, [2]>([1, 1])]; |
| 230 | tensor<int32, [2]> var_327 = const()[name = tensor<string, []>("op_327"), val = tensor<int32, [2]>([1, 1])]; |
| 231 | tensor<string, []> current_value_pad_type_0 = const()[name = tensor<string, []>("current_value_pad_type_0"), val = tensor<string, []>("custom")]; |
| 232 | tensor<int32, [4]> current_value_pad_0 = const()[name = tensor<string, []>("current_value_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 233 | tensor<fp16, [1280, 1280, 1, 1]> layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(192964736)))]; |
| 234 | tensor<fp16, [1280]> layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(196241600)))]; |
| 235 | tensor<fp16, [1, 1280, 1, 1]> current_value_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = var_327, groups = var_277, pad = current_value_pad_0, pad_type = current_value_pad_type_0, strides = var_325, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_15_cast_fp16)[name = tensor<string, []>("current_value_cast_fp16")]; |
| 236 | tensor<fp16, [1, 1280, 1, 448]> var_334_cast_fp16 = mul(x = current_key_cast_fp16, y = var_118_cast_fp16)[name = tensor<string, []>("op_334_cast_fp16")]; |
| 237 | tensor<fp16, [1, 1280, 1, 448]> var_336_cast_fp16 = mul(x = var_43_cast_fp16_1, y = var_121_cast_fp16)[name = tensor<string, []>("op_336_cast_fp16")]; |
| 238 | tensor<fp16, [1, 1280, 1, 448]> key_5_cast_fp16 = add(x = var_334_cast_fp16, y = var_336_cast_fp16)[name = tensor<string, []>("key_5_cast_fp16")]; |
| 239 | tensor<fp16, [1, 1280, 1, 448]> var_338_cast_fp16 = mul(x = current_value_cast_fp16, y = var_118_cast_fp16)[name = tensor<string, []>("op_338_cast_fp16")]; |
| 240 | tensor<fp16, [1, 1280, 1, 448]> var_340_cast_fp16 = mul(x = var_48_cast_fp16_1, y = var_121_cast_fp16)[name = tensor<string, []>("op_340_cast_fp16")]; |
| 241 | tensor<fp16, [1, 1280, 1, 448]> value_5_cast_fp16 = add(x = var_338_cast_fp16, y = var_340_cast_fp16)[name = tensor<string, []>("value_5_cast_fp16")]; |
| 242 | tensor<int32, [4]> var_343 = const()[name = tensor<string, []>("op_343"), val = tensor<int32, [4]>([1, 20, 64, -1])]; |
| 243 | tensor<fp16, [1, 20, 64, 1]> var_344_cast_fp16 = reshape(shape = var_343, x = query_5_cast_fp16)[name = tensor<string, []>("op_344_cast_fp16")]; |
| 244 | tensor<fp16, []> var_345_to_fp16 = const()[name = tensor<string, []>("op_345_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| 245 | tensor<fp16, [1, 20, 64, 1]> var_346_cast_fp16 = mul(x = var_344_cast_fp16, y = var_345_to_fp16)[name = tensor<string, []>("op_346_cast_fp16")]; |
| 246 | tensor<int32, [4]> var_347 = const()[name = tensor<string, []>("op_347"), val = tensor<int32, [4]>([1, 20, 64, -1])]; |
| 247 | tensor<fp16, [1, 20, 64, 448]> var_348_cast_fp16 = reshape(shape = var_347, x = key_5_cast_fp16)[name = tensor<string, []>("op_348_cast_fp16")]; |
| 248 | tensor<bool, []> mh_w_7_transpose_x_0 = const()[name = tensor<string, []>("mh_w_7_transpose_x_0"), val = tensor<bool, []>(true)]; |
| 249 | tensor<bool, []> mh_w_7_transpose_y_0 = const()[name = tensor<string, []>("mh_w_7_transpose_y_0"), val = tensor<bool, []>(false)]; |
| 250 | tensor<fp16, [1, 20, 1, 448]> mh_w_7_cast_fp16 = matmul(transpose_x = mh_w_7_transpose_x_0, transpose_y = mh_w_7_transpose_y_0, x = var_346_cast_fp16, y = var_348_cast_fp16)[name = tensor<string, []>("mh_w_7_cast_fp16")]; |
| 251 | tensor<fp16, [1, 20, 1, 448]> mh_w_9_cast_fp16 = add(x = mh_w_7_cast_fp16, y = var_139_cast_fp16)[name = tensor<string, []>("mh_w_9_cast_fp16")]; |
| 252 | tensor<fp16, [1, 20, 1, 448]> var_356_cast_fp16 = softmax(axis = var_270, x = mh_w_9_cast_fp16)[name = tensor<string, []>("op_356_cast_fp16")]; |
| 253 | tensor<int32, [4]> var_357 = const()[name = tensor<string, []>("op_357"), val = tensor<int32, [4]>([1, 20, 64, -1])]; |
| 254 | tensor<fp16, [1, 20, 64, 448]> var_358_cast_fp16 = reshape(shape = var_357, x = value_5_cast_fp16)[name = tensor<string, []>("op_358_cast_fp16")]; |
| 255 | tensor<bool, []> attn_5_transpose_x_0 = const()[name = tensor<string, []>("attn_5_transpose_x_0"), val = tensor<bool, []>(false)]; |
| 256 | tensor<bool, []> attn_5_transpose_y_0 = const()[name = tensor<string, []>("attn_5_transpose_y_0"), val = tensor<bool, []>(true)]; |
| 257 | tensor<fp16, [1, 20, 64, 1]> attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_358_cast_fp16, y = var_356_cast_fp16)[name = tensor<string, []>("attn_5_cast_fp16")]; |
| 258 | tensor<int32, [4]> var_361 = const()[name = tensor<string, []>("op_361"), val = tensor<int32, [4]>([1, 1280, 1, -1])]; |
| 259 | tensor<fp16, [1, 1280, 1, 1]> input_11_cast_fp16 = reshape(shape = var_361, x = attn_5_cast_fp16)[name = tensor<string, []>("input_11_cast_fp16")]; |
| 260 | tensor<int32, [2]> var_365 = const()[name = tensor<string, []>("op_365"), val = tensor<int32, [2]>([1, 1])]; |
| 261 | tensor<int32, [2]> var_367 = const()[name = tensor<string, []>("op_367"), val = tensor<int32, [2]>([1, 1])]; |
| 262 | tensor<string, []> obj_21_pad_type_0 = const()[name = tensor<string, []>("obj_21_pad_type_0"), val = tensor<string, []>("custom")]; |
| 263 | tensor<int32, [4]> obj_21_pad_0 = const()[name = tensor<string, []>("obj_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 264 | tensor<fp16, [1280, 1280, 1, 1]> layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(196244224)))]; |
| 265 | tensor<fp16, [1280]> layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(199521088)))]; |
| 266 | tensor<fp16, [1, 1280, 1, 1]> obj_21_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = var_367, groups = var_277, pad = obj_21_pad_0, pad_type = obj_21_pad_type_0, strides = var_365, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_11_cast_fp16)[name = tensor<string, []>("obj_21_cast_fp16")]; |
| 267 | tensor<fp16, [1, 1280, 1, 1]> inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = obj_21_cast_fp16)[name = tensor<string, []>("inputs_9_cast_fp16")]; |
| 268 | tensor<int32, [1]> var_377 = const()[name = tensor<string, []>("op_377"), val = tensor<int32, [1]>([1])]; |
| 269 | tensor<fp16, [1, 1, 1, 1]> channels_mean_9_cast_fp16 = reduce_mean(axes = var_377, keep_dims = var_278, x = inputs_9_cast_fp16)[name = tensor<string, []>("channels_mean_9_cast_fp16")]; |
| 270 | tensor<fp16, [1, 1280, 1, 1]> zero_mean_9_cast_fp16 = sub(x = inputs_9_cast_fp16, y = channels_mean_9_cast_fp16)[name = tensor<string, []>("zero_mean_9_cast_fp16")]; |
| 271 | tensor<fp16, [1, 1280, 1, 1]> zero_mean_sq_9_cast_fp16 = mul(x = zero_mean_9_cast_fp16, y = zero_mean_9_cast_fp16)[name = tensor<string, []>("zero_mean_sq_9_cast_fp16")]; |
| 272 | tensor<int32, [1]> var_381 = const()[name = tensor<string, []>("op_381"), val = tensor<int32, [1]>([1])]; |
| 273 | tensor<fp16, [1, 1, 1, 1]> var_382_cast_fp16 = reduce_mean(axes = var_381, keep_dims = var_278, x = zero_mean_sq_9_cast_fp16)[name = tensor<string, []>("op_382_cast_fp16")]; |
| 274 | tensor<fp16, []> var_383_to_fp16 = const()[name = tensor<string, []>("op_383_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 275 | tensor<fp16, [1, 1, 1, 1]> var_384_cast_fp16 = add(x = var_382_cast_fp16, y = var_383_to_fp16)[name = tensor<string, []>("op_384_cast_fp16")]; |
| 276 | tensor<fp16, []> denom_9_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_9_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)]; |
| 277 | tensor<fp16, [1, 1, 1, 1]> denom_9_cast_fp16 = rsqrt(epsilon = denom_9_epsilon_0_to_fp16, x = var_384_cast_fp16)[name = tensor<string, []>("denom_9_cast_fp16")]; |
| 278 | tensor<fp16, [1, 1280, 1, 1]> out_9_cast_fp16 = mul(x = zero_mean_9_cast_fp16, y = denom_9_cast_fp16)[name = tensor<string, []>("out_9_cast_fp16")]; |
| 279 | tensor<fp16, [1280]> obj_23_gamma_0_to_fp16 = const()[name = tensor<string, []>("obj_23_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(199523712)))]; |
| 280 | tensor<fp16, [1280]> obj_23_beta_0_to_fp16 = const()[name = tensor<string, []>("obj_23_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(199526336)))]; |
| 281 | tensor<fp16, []> obj_23_epsilon_0_to_fp16 = const()[name = tensor<string, []>("obj_23_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 282 | tensor<fp16, [1, 1280, 1, 1]> obj_23_cast_fp16 = batch_norm(beta = obj_23_beta_0_to_fp16, epsilon = obj_23_epsilon_0_to_fp16, gamma = obj_23_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_9_cast_fp16)[name = tensor<string, []>("obj_23_cast_fp16")]; |
| 283 | tensor<int32, [2]> var_399 = const()[name = tensor<string, []>("op_399"), val = tensor<int32, [2]>([1, 1])]; |
| 284 | tensor<int32, [2]> var_401 = const()[name = tensor<string, []>("op_401"), val = tensor<int32, [2]>([1, 1])]; |
| 285 | tensor<string, []> query_pad_type_0 = const()[name = tensor<string, []>("query_pad_type_0"), val = tensor<string, []>("custom")]; |
| 286 | tensor<int32, [4]> query_pad_0 = const()[name = tensor<string, []>("query_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 287 | tensor<fp16, [1280, 1280, 1, 1]> layers_1_encoder_attn_q_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_encoder_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(199528960)))]; |
| 288 | tensor<fp16, [1280]> layers_1_encoder_attn_q_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_encoder_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(202805824)))]; |
| 289 | tensor<fp16, [1, 1280, 1, 1]> query_cast_fp16 = conv(bias = layers_1_encoder_attn_q_proj_bias_to_fp16, dilations = var_401, groups = var_277, pad = query_pad_0, pad_type = query_pad_type_0, strides = var_399, weight = layers_1_encoder_attn_q_proj_weight_to_fp16, x = obj_23_cast_fp16)[name = tensor<string, []>("query_cast_fp16")]; |
| 290 | tensor<int32, [2]> var_405 = const()[name = tensor<string, []>("op_405"), val = tensor<int32, [2]>([1, 1])]; |
| 291 | tensor<int32, [2]> var_407 = const()[name = tensor<string, []>("op_407"), val = tensor<int32, [2]>([1, 1])]; |
| 292 | tensor<string, []> key_pad_type_0 = const()[name = tensor<string, []>("key_pad_type_0"), val = tensor<string, []>("custom")]; |
| 293 | tensor<int32, [4]> key_pad_0 = const()[name = tensor<string, []>("key_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 294 | tensor<fp16, [1280, 1280, 1, 1]> layers_1_encoder_attn_k_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_encoder_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(202808448)))]; |
| 295 | tensor<fp16, [1, 1280, 1, 1500]> key_cast_fp16 = conv(dilations = var_407, groups = var_277, pad = key_pad_0, pad_type = key_pad_type_0, strides = var_405, weight = layers_1_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("key_cast_fp16")]; |
| 296 | tensor<int32, [2]> var_412 = const()[name = tensor<string, []>("op_412"), val = tensor<int32, [2]>([1, 1])]; |
| 297 | tensor<int32, [2]> var_414 = const()[name = tensor<string, []>("op_414"), val = tensor<int32, [2]>([1, 1])]; |
| 298 | tensor<string, []> value_pad_type_0 = const()[name = tensor<string, []>("value_pad_type_0"), val = tensor<string, []>("custom")]; |
| 299 | tensor<int32, [4]> value_pad_0 = const()[name = tensor<string, []>("value_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 300 | tensor<fp16, [1280, 1280, 1, 1]> layers_1_encoder_attn_v_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_encoder_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(206085312)))]; |
| 301 | tensor<fp16, [1280]> layers_1_encoder_attn_v_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(209362176)))]; |
| 302 | tensor<fp16, [1, 1280, 1, 1500]> value_cast_fp16 = conv(bias = layers_1_encoder_attn_v_proj_bias_to_fp16, dilations = var_414, groups = var_277, pad = value_pad_0, pad_type = value_pad_type_0, strides = var_412, weight = layers_1_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = tensor<string, []>("value_cast_fp16")]; |
| 303 | tensor<int32, [4]> var_418 = const()[name = tensor<string, []>("op_418"), val = tensor<int32, [4]>([1, 20, 64, -1])]; |
| 304 | tensor<fp16, [1, 20, 64, 1]> var_419_cast_fp16 = reshape(shape = var_418, x = query_cast_fp16)[name = tensor<string, []>("op_419_cast_fp16")]; |
| 305 | tensor<fp16, []> var_420_to_fp16 = const()[name = tensor<string, []>("op_420_to_fp16"), val = tensor<fp16, []>(0x1p-3)]; |
| 306 | tensor<fp16, [1, 20, 64, 1]> var_421_cast_fp16 = mul(x = var_419_cast_fp16, y = var_420_to_fp16)[name = tensor<string, []>("op_421_cast_fp16")]; |
| 307 | tensor<int32, [4]> var_422 = const()[name = tensor<string, []>("op_422"), val = tensor<int32, [4]>([1, 20, 64, -1])]; |
| 308 | tensor<fp16, [1, 20, 64, 1500]> var_423_cast_fp16 = reshape(shape = var_422, x = key_cast_fp16)[name = tensor<string, []>("op_423_cast_fp16")]; |
| 309 | tensor<bool, []> mh_w_transpose_x_0 = const()[name = tensor<string, []>("mh_w_transpose_x_0"), val = tensor<bool, []>(true)]; |
| 310 | tensor<bool, []> mh_w_transpose_y_0 = const()[name = tensor<string, []>("mh_w_transpose_y_0"), val = tensor<bool, []>(false)]; |
| 311 | tensor<fp16, [1, 20, 1, 1500]> mh_w_cast_fp16 = matmul(transpose_x = mh_w_transpose_x_0, transpose_y = mh_w_transpose_y_0, x = var_421_cast_fp16, y = var_423_cast_fp16)[name = tensor<string, []>("mh_w_cast_fp16")]; |
| 312 | tensor<fp16, [1, 20, 1, 1500]> obj_27_cast_fp16 = softmax(axis = var_270, x = mh_w_cast_fp16)[name = tensor<string, []>("obj_27_cast_fp16")]; |
| 313 | tensor<int32, [4]> var_427 = const()[name = tensor<string, []>("op_427"), val = tensor<int32, [4]>([1, 20, 64, -1])]; |
| 314 | tensor<fp16, [1, 20, 64, 1500]> var_428_cast_fp16 = reshape(shape = var_427, x = value_cast_fp16)[name = tensor<string, []>("op_428_cast_fp16")]; |
| 315 | tensor<bool, []> attn_transpose_x_0 = const()[name = tensor<string, []>("attn_transpose_x_0"), val = tensor<bool, []>(false)]; |
| 316 | tensor<bool, []> attn_transpose_y_0 = const()[name = tensor<string, []>("attn_transpose_y_0"), val = tensor<bool, []>(true)]; |
| 317 | tensor<fp16, [1, 20, 64, 1]> attn_cast_fp16 = matmul(transpose_x = attn_transpose_x_0, transpose_y = attn_transpose_y_0, x = var_428_cast_fp16, y = obj_27_cast_fp16)[name = tensor<string, []>("attn_cast_fp16")]; |
| 318 | tensor<int32, [4]> var_431 = const()[name = tensor<string, []>("op_431"), val = tensor<int32, [4]>([1, 1280, 1, -1])]; |
| 319 | tensor<fp16, [1, 1280, 1, 1]> input_13_cast_fp16 = reshape(shape = var_431, x = attn_cast_fp16)[name = tensor<string, []>("input_13_cast_fp16")]; |
| 320 | tensor<int32, [2]> var_435 = const()[name = tensor<string, []>("op_435"), val = tensor<int32, [2]>([1, 1])]; |
| 321 | tensor<int32, [2]> var_437 = const()[name = tensor<string, []>("op_437"), val = tensor<int32, [2]>([1, 1])]; |
| 322 | tensor<string, []> obj_25_pad_type_0 = const()[name = tensor<string, []>("obj_25_pad_type_0"), val = tensor<string, []>("custom")]; |
| 323 | tensor<int32, [4]> obj_25_pad_0 = const()[name = tensor<string, []>("obj_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 324 | tensor<fp16, [1280, 1280, 1, 1]> layers_1_encoder_attn_o_proj_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_encoder_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1280, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(209364800)))]; |
| 325 | tensor<fp16, [1280]> layers_1_encoder_attn_o_proj_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_encoder_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(212641664)))]; |
| 326 | tensor<fp16, [1, 1280, 1, 1]> obj_25_cast_fp16 = conv(bias = layers_1_encoder_attn_o_proj_bias_to_fp16, dilations = var_437, groups = var_277, pad = obj_25_pad_0, pad_type = obj_25_pad_type_0, strides = var_435, weight = layers_1_encoder_attn_o_proj_weight_to_fp16, x = input_13_cast_fp16)[name = tensor<string, []>("obj_25_cast_fp16")]; |
| 327 | tensor<fp16, [1, 1280, 1, 1]> inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_25_cast_fp16)[name = tensor<string, []>("inputs_11_cast_fp16")]; |
| 328 | tensor<int32, [1]> var_446 = const()[name = tensor<string, []>("op_446"), val = tensor<int32, [1]>([1])]; |
| 329 | tensor<fp16, [1, 1, 1, 1]> channels_mean_11_cast_fp16 = reduce_mean(axes = var_446, keep_dims = var_278, x = inputs_11_cast_fp16)[name = tensor<string, []>("channels_mean_11_cast_fp16")]; |
| 330 | tensor<fp16, [1, 1280, 1, 1]> zero_mean_11_cast_fp16 = sub(x = inputs_11_cast_fp16, y = channels_mean_11_cast_fp16)[name = tensor<string, []>("zero_mean_11_cast_fp16")]; |
| 331 | tensor<fp16, [1, 1280, 1, 1]> zero_mean_sq_11_cast_fp16 = mul(x = zero_mean_11_cast_fp16, y = zero_mean_11_cast_fp16)[name = tensor<string, []>("zero_mean_sq_11_cast_fp16")]; |
| 332 | tensor<int32, [1]> var_450 = const()[name = tensor<string, []>("op_450"), val = tensor<int32, [1]>([1])]; |
| 333 | tensor<fp16, [1, 1, 1, 1]> var_451_cast_fp16 = reduce_mean(axes = var_450, keep_dims = var_278, x = zero_mean_sq_11_cast_fp16)[name = tensor<string, []>("op_451_cast_fp16")]; |
| 334 | tensor<fp16, []> var_452_to_fp16 = const()[name = tensor<string, []>("op_452_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 335 | tensor<fp16, [1, 1, 1, 1]> var_453_cast_fp16 = add(x = var_451_cast_fp16, y = var_452_to_fp16)[name = tensor<string, []>("op_453_cast_fp16")]; |
| 336 | tensor<fp16, []> denom_11_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_11_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)]; |
| 337 | tensor<fp16, [1, 1, 1, 1]> denom_11_cast_fp16 = rsqrt(epsilon = denom_11_epsilon_0_to_fp16, x = var_453_cast_fp16)[name = tensor<string, []>("denom_11_cast_fp16")]; |
| 338 | tensor<fp16, [1, 1280, 1, 1]> out_11_cast_fp16 = mul(x = zero_mean_11_cast_fp16, y = denom_11_cast_fp16)[name = tensor<string, []>("out_11_cast_fp16")]; |
| 339 | tensor<fp16, [1280]> input_15_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_15_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(212644288)))]; |
| 340 | tensor<fp16, [1280]> input_15_beta_0_to_fp16 = const()[name = tensor<string, []>("input_15_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(212646912)))]; |
| 341 | tensor<fp16, []> input_15_epsilon_0_to_fp16 = const()[name = tensor<string, []>("input_15_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 342 | tensor<fp16, [1, 1280, 1, 1]> input_15_cast_fp16 = batch_norm(beta = input_15_beta_0_to_fp16, epsilon = input_15_epsilon_0_to_fp16, gamma = input_15_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_11_cast_fp16)[name = tensor<string, []>("input_15_cast_fp16")]; |
| 343 | tensor<int32, [2]> var_464 = const()[name = tensor<string, []>("op_464"), val = tensor<int32, [2]>([1, 1])]; |
| 344 | tensor<int32, [2]> var_466 = const()[name = tensor<string, []>("op_466"), val = tensor<int32, [2]>([1, 1])]; |
| 345 | tensor<string, []> input_17_pad_type_0 = const()[name = tensor<string, []>("input_17_pad_type_0"), val = tensor<string, []>("custom")]; |
| 346 | tensor<int32, [4]> input_17_pad_0 = const()[name = tensor<string, []>("input_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 347 | tensor<fp16, [5120, 1280, 1, 1]> layers_1_fc1_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_fc1_weight_to_fp16"), val = tensor<fp16, [5120, 1280, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(212649536)))]; |
| 348 | tensor<fp16, [5120]> layers_1_fc1_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_fc1_bias_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(225756800)))]; |
| 349 | tensor<fp16, [1, 5120, 1, 1]> input_17_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = var_466, groups = var_277, pad = input_17_pad_0, pad_type = input_17_pad_type_0, strides = var_464, weight = layers_1_fc1_weight_to_fp16, x = input_15_cast_fp16)[name = tensor<string, []>("input_17_cast_fp16")]; |
| 350 | tensor<string, []> input_mode_0 = const()[name = tensor<string, []>("input_mode_0"), val = tensor<string, []>("EXACT")]; |
| 351 | tensor<fp16, [1, 5120, 1, 1]> input_cast_fp16 = gelu(mode = input_mode_0, x = input_17_cast_fp16)[name = tensor<string, []>("input_cast_fp16")]; |
| 352 | tensor<int32, [2]> var_472 = const()[name = tensor<string, []>("op_472"), val = tensor<int32, [2]>([1, 1])]; |
| 353 | tensor<int32, [2]> var_474 = const()[name = tensor<string, []>("op_474"), val = tensor<int32, [2]>([1, 1])]; |
| 354 | tensor<string, []> hidden_states_5_pad_type_0 = const()[name = tensor<string, []>("hidden_states_5_pad_type_0"), val = tensor<string, []>("custom")]; |
| 355 | tensor<int32, [4]> hidden_states_5_pad_0 = const()[name = tensor<string, []>("hidden_states_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 356 | tensor<fp16, [1280, 5120, 1, 1]> layers_1_fc2_weight_to_fp16 = const()[name = tensor<string, []>("layers_1_fc2_weight_to_fp16"), val = tensor<fp16, [1280, 5120, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(225767104)))]; |
| 357 | tensor<fp16, [1280]> layers_1_fc2_bias_to_fp16 = const()[name = tensor<string, []>("layers_1_fc2_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(238874368)))]; |
| 358 | tensor<fp16, [1, 1280, 1, 1]> hidden_states_5_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = var_474, groups = var_277, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = var_472, weight = layers_1_fc2_weight_to_fp16, x = input_cast_fp16)[name = tensor<string, []>("hidden_states_5_cast_fp16")]; |
| 359 | tensor<fp16, [1, 1280, 1, 1]> inputs_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_5_cast_fp16)[name = tensor<string, []>("inputs_cast_fp16")]; |
| 360 | tensor<bool, []> var_485 = const()[name = tensor<string, []>("op_485"), val = tensor<bool, []>(true)]; |
| 361 | tensor<int32, [1]> var_489 = const()[name = tensor<string, []>("op_489"), val = tensor<int32, [1]>([1])]; |
| 362 | tensor<fp16, [1, 1, 1, 1]> channels_mean_cast_fp16 = reduce_mean(axes = var_489, keep_dims = var_485, x = inputs_cast_fp16)[name = tensor<string, []>("channels_mean_cast_fp16")]; |
| 363 | tensor<fp16, [1, 1280, 1, 1]> zero_mean_cast_fp16 = sub(x = inputs_cast_fp16, y = channels_mean_cast_fp16)[name = tensor<string, []>("zero_mean_cast_fp16")]; |
| 364 | tensor<fp16, [1, 1280, 1, 1]> zero_mean_sq_cast_fp16 = mul(x = zero_mean_cast_fp16, y = zero_mean_cast_fp16)[name = tensor<string, []>("zero_mean_sq_cast_fp16")]; |
| 365 | tensor<int32, [1]> var_493 = const()[name = tensor<string, []>("op_493"), val = tensor<int32, [1]>([1])]; |
| 366 | tensor<fp16, [1, 1, 1, 1]> var_494_cast_fp16 = reduce_mean(axes = var_493, keep_dims = var_485, x = zero_mean_sq_cast_fp16)[name = tensor<string, []>("op_494_cast_fp16")]; |
| 367 | tensor<fp16, []> var_495_to_fp16 = const()[name = tensor<string, []>("op_495_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 368 | tensor<fp16, [1, 1, 1, 1]> var_496_cast_fp16 = add(x = var_494_cast_fp16, y = var_495_to_fp16)[name = tensor<string, []>("op_496_cast_fp16")]; |
| 369 | tensor<fp16, []> denom_epsilon_0_to_fp16 = const()[name = tensor<string, []>("denom_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1p-24)]; |
| 370 | tensor<fp16, [1, 1, 1, 1]> denom_cast_fp16 = rsqrt(epsilon = denom_epsilon_0_to_fp16, x = var_496_cast_fp16)[name = tensor<string, []>("denom_cast_fp16")]; |
| 371 | tensor<fp16, [1, 1280, 1, 1]> out_cast_fp16 = mul(x = zero_mean_cast_fp16, y = denom_cast_fp16)[name = tensor<string, []>("out_cast_fp16")]; |
| 372 | tensor<fp16, [1280]> hidden_states_gamma_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_gamma_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(238876992)))]; |
| 373 | tensor<fp16, [1280]> hidden_states_beta_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_beta_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(238879616)))]; |
| 374 | tensor<fp16, []> hidden_states_epsilon_0_to_fp16 = const()[name = tensor<string, []>("hidden_states_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| 375 | tensor<fp16, [1, 1280, 1, 1]> hidden_states_cast_fp16 = batch_norm(beta = hidden_states_beta_0_to_fp16, epsilon = hidden_states_epsilon_0_to_fp16, gamma = hidden_states_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_cast_fp16)[name = tensor<string, []>("hidden_states_cast_fp16")]; |
| 376 | tensor<int32, [1]> var_506_axes_0 = const()[name = tensor<string, []>("op_506_axes_0"), val = tensor<int32, [1]>([2])]; |
| 377 | tensor<fp16, [1, 1280, 1]> var_506_cast_fp16 = squeeze(axes = var_506_axes_0, x = hidden_states_cast_fp16)[name = tensor<string, []>("op_506_cast_fp16")]; |
| 378 | tensor<int32, [3]> var_509_perm_0 = const()[name = tensor<string, []>("op_509_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; |
| 379 | tensor<fp16, [51866]> linear_0_bias_0_to_fp16 = const()[name = tensor<string, []>("linear_0_bias_0_to_fp16"), val = tensor<fp16, [51866]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(238882240)))]; |
| 380 | tensor<fp16, [1, 1, 1280]> transpose_0 = transpose(perm = var_509_perm_0, x = var_506_cast_fp16)[name = tensor<string, []>("transpose_0")]; |
| 381 | tensor<fp16, [1, 1, 51866]> logits = linear(bias = linear_0_bias_0_to_fp16, weight = embed_tokens_weight_to_fp16, x = transpose_0)[name = tensor<string, []>("linear_0_cast_fp16")]; |
| 382 | tensor<int32, []> var_513 = const()[name = tensor<string, []>("op_513"), val = tensor<int32, []>(1)]; |
| 383 | tensor<bool, []> obj_31_interleave_0 = const()[name = tensor<string, []>("obj_31_interleave_0"), val = tensor<bool, []>(false)]; |
| 384 | tensor<fp16, [1, 2560, 1, 1]> key_cache_updates = concat(axis = var_513, interleave = obj_31_interleave_0, values = (current_key_1_cast_fp16, current_key_cast_fp16))[name = tensor<string, []>("obj_31_cast_fp16")]; |
| 385 | tensor<int32, []> var_516 = const()[name = tensor<string, []>("op_516"), val = tensor<int32, []>(1)]; |
| 386 | tensor<bool, []> obj_33_interleave_0 = const()[name = tensor<string, []>("obj_33_interleave_0"), val = tensor<bool, []>(false)]; |
| 387 | tensor<fp16, [1, 2560, 1, 1]> value_cache_updates = concat(axis = var_516, interleave = obj_33_interleave_0, values = (current_value_1_cast_fp16, current_value_cast_fp16))[name = tensor<string, []>("obj_33_cast_fp16")]; |
| 388 | tensor<int32, [4]> var_527_begin_0 = const()[name = tensor<string, []>("op_527_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 389 | tensor<int32, [4]> var_527_end_0 = const()[name = tensor<string, []>("op_527_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| 390 | tensor<bool, [4]> var_527_end_mask_0 = const()[name = tensor<string, []>("op_527_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| 391 | tensor<fp16, [1, 1, 1, 1500]> var_527_cast_fp16 = slice_by_index(begin = var_527_begin_0, end = var_527_end_0, end_mask = var_527_end_mask_0, x = obj_27_cast_fp16)[name = tensor<string, []>("op_527_cast_fp16")]; |
| 392 | tensor<int32, [4]> var_530_begin_0 = const()[name = tensor<string, []>("op_530_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 393 | tensor<int32, [4]> var_530_end_0 = const()[name = tensor<string, []>("op_530_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| 394 | tensor<bool, [4]> var_530_end_mask_0 = const()[name = tensor<string, []>("op_530_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| 395 | tensor<bool, [4]> var_530_squeeze_mask_0 = const()[name = tensor<string, []>("op_530_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| 396 | tensor<fp16, [1, 1, 1500]> var_530_cast_fp16 = slice_by_index(begin = var_530_begin_0, end = var_530_end_0, end_mask = var_530_end_mask_0, squeeze_mask = var_530_squeeze_mask_0, x = var_527_cast_fp16)[name = tensor<string, []>("op_530_cast_fp16")]; |
| 397 | tensor<int32, [4]> var_545_begin_0 = const()[name = tensor<string, []>("op_545_begin_0"), val = tensor<int32, [4]>([0, 1, 0, 0])]; |
| 398 | tensor<int32, [4]> var_545_end_0 = const()[name = tensor<string, []>("op_545_end_0"), val = tensor<int32, [4]>([1, 2, 1, 1500])]; |
| 399 | tensor<bool, [4]> var_545_end_mask_0 = const()[name = tensor<string, []>("op_545_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| 400 | tensor<fp16, [1, 1, 1, 1500]> var_545_cast_fp16 = slice_by_index(begin = var_545_begin_0, end = var_545_end_0, end_mask = var_545_end_mask_0, x = obj_27_cast_fp16)[name = tensor<string, []>("op_545_cast_fp16")]; |
| 401 | tensor<int32, [4]> var_548_begin_0 = const()[name = tensor<string, []>("op_548_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 402 | tensor<int32, [4]> var_548_end_0 = const()[name = tensor<string, []>("op_548_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| 403 | tensor<bool, [4]> var_548_end_mask_0 = const()[name = tensor<string, []>("op_548_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| 404 | tensor<bool, [4]> var_548_squeeze_mask_0 = const()[name = tensor<string, []>("op_548_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| 405 | tensor<fp16, [1, 1, 1500]> var_548_cast_fp16 = slice_by_index(begin = var_548_begin_0, end = var_548_end_0, end_mask = var_548_end_mask_0, squeeze_mask = var_548_squeeze_mask_0, x = var_545_cast_fp16)[name = tensor<string, []>("op_548_cast_fp16")]; |
| 406 | tensor<int32, [4]> var_563_begin_0 = const()[name = tensor<string, []>("op_563_begin_0"), val = tensor<int32, [4]>([0, 2, 0, 0])]; |
| 407 | tensor<int32, [4]> var_563_end_0 = const()[name = tensor<string, []>("op_563_end_0"), val = tensor<int32, [4]>([1, 3, 1, 1500])]; |
| 408 | tensor<bool, [4]> var_563_end_mask_0 = const()[name = tensor<string, []>("op_563_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| 409 | tensor<fp16, [1, 1, 1, 1500]> var_563_cast_fp16 = slice_by_index(begin = var_563_begin_0, end = var_563_end_0, end_mask = var_563_end_mask_0, x = obj_27_cast_fp16)[name = tensor<string, []>("op_563_cast_fp16")]; |
| 410 | tensor<int32, [4]> var_566_begin_0 = const()[name = tensor<string, []>("op_566_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 411 | tensor<int32, [4]> var_566_end_0 = const()[name = tensor<string, []>("op_566_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| 412 | tensor<bool, [4]> var_566_end_mask_0 = const()[name = tensor<string, []>("op_566_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| 413 | tensor<bool, [4]> var_566_squeeze_mask_0 = const()[name = tensor<string, []>("op_566_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| 414 | tensor<fp16, [1, 1, 1500]> var_566_cast_fp16 = slice_by_index(begin = var_566_begin_0, end = var_566_end_0, end_mask = var_566_end_mask_0, squeeze_mask = var_566_squeeze_mask_0, x = var_563_cast_fp16)[name = tensor<string, []>("op_566_cast_fp16")]; |
| 415 | tensor<int32, [4]> var_581_begin_0 = const()[name = tensor<string, []>("op_581_begin_0"), val = tensor<int32, [4]>([0, 3, 0, 0])]; |
| 416 | tensor<int32, [4]> var_581_end_0 = const()[name = tensor<string, []>("op_581_end_0"), val = tensor<int32, [4]>([1, 4, 1, 1500])]; |
| 417 | tensor<bool, [4]> var_581_end_mask_0 = const()[name = tensor<string, []>("op_581_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| 418 | tensor<fp16, [1, 1, 1, 1500]> var_581_cast_fp16 = slice_by_index(begin = var_581_begin_0, end = var_581_end_0, end_mask = var_581_end_mask_0, x = obj_27_cast_fp16)[name = tensor<string, []>("op_581_cast_fp16")]; |
| 419 | tensor<int32, [4]> var_584_begin_0 = const()[name = tensor<string, []>("op_584_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 420 | tensor<int32, [4]> var_584_end_0 = const()[name = tensor<string, []>("op_584_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| 421 | tensor<bool, [4]> var_584_end_mask_0 = const()[name = tensor<string, []>("op_584_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| 422 | tensor<bool, [4]> var_584_squeeze_mask_0 = const()[name = tensor<string, []>("op_584_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| 423 | tensor<fp16, [1, 1, 1500]> var_584_cast_fp16 = slice_by_index(begin = var_584_begin_0, end = var_584_end_0, end_mask = var_584_end_mask_0, squeeze_mask = var_584_squeeze_mask_0, x = var_581_cast_fp16)[name = tensor<string, []>("op_584_cast_fp16")]; |
| 424 | tensor<int32, [4]> var_599_begin_0 = const()[name = tensor<string, []>("op_599_begin_0"), val = tensor<int32, [4]>([0, 4, 0, 0])]; |
| 425 | tensor<int32, [4]> var_599_end_0 = const()[name = tensor<string, []>("op_599_end_0"), val = tensor<int32, [4]>([1, 5, 1, 1500])]; |
| 426 | tensor<bool, [4]> var_599_end_mask_0 = const()[name = tensor<string, []>("op_599_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| 427 | tensor<fp16, [1, 1, 1, 1500]> var_599_cast_fp16 = slice_by_index(begin = var_599_begin_0, end = var_599_end_0, end_mask = var_599_end_mask_0, x = obj_27_cast_fp16)[name = tensor<string, []>("op_599_cast_fp16")]; |
| 428 | tensor<int32, [4]> var_602_begin_0 = const()[name = tensor<string, []>("op_602_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 429 | tensor<int32, [4]> var_602_end_0 = const()[name = tensor<string, []>("op_602_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| 430 | tensor<bool, [4]> var_602_end_mask_0 = const()[name = tensor<string, []>("op_602_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| 431 | tensor<bool, [4]> var_602_squeeze_mask_0 = const()[name = tensor<string, []>("op_602_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| 432 | tensor<fp16, [1, 1, 1500]> var_602_cast_fp16 = slice_by_index(begin = var_602_begin_0, end = var_602_end_0, end_mask = var_602_end_mask_0, squeeze_mask = var_602_squeeze_mask_0, x = var_599_cast_fp16)[name = tensor<string, []>("op_602_cast_fp16")]; |
| 433 | tensor<int32, [4]> var_617_begin_0 = const()[name = tensor<string, []>("op_617_begin_0"), val = tensor<int32, [4]>([0, 5, 0, 0])]; |
| 434 | tensor<int32, [4]> var_617_end_0 = const()[name = tensor<string, []>("op_617_end_0"), val = tensor<int32, [4]>([1, 6, 1, 1500])]; |
| 435 | tensor<bool, [4]> var_617_end_mask_0 = const()[name = tensor<string, []>("op_617_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| 436 | tensor<fp16, [1, 1, 1, 1500]> var_617_cast_fp16 = slice_by_index(begin = var_617_begin_0, end = var_617_end_0, end_mask = var_617_end_mask_0, x = obj_27_cast_fp16)[name = tensor<string, []>("op_617_cast_fp16")]; |
| 437 | tensor<int32, [4]> var_620_begin_0 = const()[name = tensor<string, []>("op_620_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 438 | tensor<int32, [4]> var_620_end_0 = const()[name = tensor<string, []>("op_620_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| 439 | tensor<bool, [4]> var_620_end_mask_0 = const()[name = tensor<string, []>("op_620_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| 440 | tensor<bool, [4]> var_620_squeeze_mask_0 = const()[name = tensor<string, []>("op_620_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| 441 | tensor<fp16, [1, 1, 1500]> var_620_cast_fp16 = slice_by_index(begin = var_620_begin_0, end = var_620_end_0, end_mask = var_620_end_mask_0, squeeze_mask = var_620_squeeze_mask_0, x = var_617_cast_fp16)[name = tensor<string, []>("op_620_cast_fp16")]; |
| 442 | tensor<int32, [4]> var_635_begin_0 = const()[name = tensor<string, []>("op_635_begin_0"), val = tensor<int32, [4]>([0, 6, 0, 0])]; |
| 443 | tensor<int32, [4]> var_635_end_0 = const()[name = tensor<string, []>("op_635_end_0"), val = tensor<int32, [4]>([1, 7, 1, 1500])]; |
| 444 | tensor<bool, [4]> var_635_end_mask_0 = const()[name = tensor<string, []>("op_635_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| 445 | tensor<fp16, [1, 1, 1, 1500]> var_635_cast_fp16 = slice_by_index(begin = var_635_begin_0, end = var_635_end_0, end_mask = var_635_end_mask_0, x = obj_27_cast_fp16)[name = tensor<string, []>("op_635_cast_fp16")]; |
| 446 | tensor<int32, [4]> var_638_begin_0 = const()[name = tensor<string, []>("op_638_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 447 | tensor<int32, [4]> var_638_end_0 = const()[name = tensor<string, []>("op_638_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| 448 | tensor<bool, [4]> var_638_end_mask_0 = const()[name = tensor<string, []>("op_638_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| 449 | tensor<bool, [4]> var_638_squeeze_mask_0 = const()[name = tensor<string, []>("op_638_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| 450 | tensor<fp16, [1, 1, 1500]> var_638_cast_fp16 = slice_by_index(begin = var_638_begin_0, end = var_638_end_0, end_mask = var_638_end_mask_0, squeeze_mask = var_638_squeeze_mask_0, x = var_635_cast_fp16)[name = tensor<string, []>("op_638_cast_fp16")]; |
| 451 | tensor<int32, [4]> var_653_begin_0 = const()[name = tensor<string, []>("op_653_begin_0"), val = tensor<int32, [4]>([0, 7, 0, 0])]; |
| 452 | tensor<int32, [4]> var_653_end_0 = const()[name = tensor<string, []>("op_653_end_0"), val = tensor<int32, [4]>([1, 8, 1, 1500])]; |
| 453 | tensor<bool, [4]> var_653_end_mask_0 = const()[name = tensor<string, []>("op_653_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| 454 | tensor<fp16, [1, 1, 1, 1500]> var_653_cast_fp16 = slice_by_index(begin = var_653_begin_0, end = var_653_end_0, end_mask = var_653_end_mask_0, x = obj_27_cast_fp16)[name = tensor<string, []>("op_653_cast_fp16")]; |
| 455 | tensor<int32, [4]> var_656_begin_0 = const()[name = tensor<string, []>("op_656_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 456 | tensor<int32, [4]> var_656_end_0 = const()[name = tensor<string, []>("op_656_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| 457 | tensor<bool, [4]> var_656_end_mask_0 = const()[name = tensor<string, []>("op_656_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| 458 | tensor<bool, [4]> var_656_squeeze_mask_0 = const()[name = tensor<string, []>("op_656_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| 459 | tensor<fp16, [1, 1, 1500]> var_656_cast_fp16 = slice_by_index(begin = var_656_begin_0, end = var_656_end_0, end_mask = var_656_end_mask_0, squeeze_mask = var_656_squeeze_mask_0, x = var_653_cast_fp16)[name = tensor<string, []>("op_656_cast_fp16")]; |
| 460 | tensor<int32, [4]> var_671_begin_0 = const()[name = tensor<string, []>("op_671_begin_0"), val = tensor<int32, [4]>([0, 8, 0, 0])]; |
| 461 | tensor<int32, [4]> var_671_end_0 = const()[name = tensor<string, []>("op_671_end_0"), val = tensor<int32, [4]>([1, 9, 1, 1500])]; |
| 462 | tensor<bool, [4]> var_671_end_mask_0 = const()[name = tensor<string, []>("op_671_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| 463 | tensor<fp16, [1, 1, 1, 1500]> var_671_cast_fp16 = slice_by_index(begin = var_671_begin_0, end = var_671_end_0, end_mask = var_671_end_mask_0, x = obj_27_cast_fp16)[name = tensor<string, []>("op_671_cast_fp16")]; |
| 464 | tensor<int32, [4]> var_674_begin_0 = const()[name = tensor<string, []>("op_674_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 465 | tensor<int32, [4]> var_674_end_0 = const()[name = tensor<string, []>("op_674_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| 466 | tensor<bool, [4]> var_674_end_mask_0 = const()[name = tensor<string, []>("op_674_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| 467 | tensor<bool, [4]> var_674_squeeze_mask_0 = const()[name = tensor<string, []>("op_674_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| 468 | tensor<fp16, [1, 1, 1500]> var_674_cast_fp16 = slice_by_index(begin = var_674_begin_0, end = var_674_end_0, end_mask = var_674_end_mask_0, squeeze_mask = var_674_squeeze_mask_0, x = var_671_cast_fp16)[name = tensor<string, []>("op_674_cast_fp16")]; |
| 469 | tensor<int32, [4]> var_689_begin_0 = const()[name = tensor<string, []>("op_689_begin_0"), val = tensor<int32, [4]>([0, 9, 0, 0])]; |
| 470 | tensor<int32, [4]> var_689_end_0 = const()[name = tensor<string, []>("op_689_end_0"), val = tensor<int32, [4]>([1, 10, 1, 1500])]; |
| 471 | tensor<bool, [4]> var_689_end_mask_0 = const()[name = tensor<string, []>("op_689_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| 472 | tensor<fp16, [1, 1, 1, 1500]> var_689_cast_fp16 = slice_by_index(begin = var_689_begin_0, end = var_689_end_0, end_mask = var_689_end_mask_0, x = obj_27_cast_fp16)[name = tensor<string, []>("op_689_cast_fp16")]; |
| 473 | tensor<int32, [4]> var_692_begin_0 = const()[name = tensor<string, []>("op_692_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 474 | tensor<int32, [4]> var_692_end_0 = const()[name = tensor<string, []>("op_692_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| 475 | tensor<bool, [4]> var_692_end_mask_0 = const()[name = tensor<string, []>("op_692_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| 476 | tensor<bool, [4]> var_692_squeeze_mask_0 = const()[name = tensor<string, []>("op_692_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| 477 | tensor<fp16, [1, 1, 1500]> var_692_cast_fp16 = slice_by_index(begin = var_692_begin_0, end = var_692_end_0, end_mask = var_692_end_mask_0, squeeze_mask = var_692_squeeze_mask_0, x = var_689_cast_fp16)[name = tensor<string, []>("op_692_cast_fp16")]; |
| 478 | tensor<int32, [4]> var_707_begin_0 = const()[name = tensor<string, []>("op_707_begin_0"), val = tensor<int32, [4]>([0, 10, 0, 0])]; |
| 479 | tensor<int32, [4]> var_707_end_0 = const()[name = tensor<string, []>("op_707_end_0"), val = tensor<int32, [4]>([1, 11, 1, 1500])]; |
| 480 | tensor<bool, [4]> var_707_end_mask_0 = const()[name = tensor<string, []>("op_707_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| 481 | tensor<fp16, [1, 1, 1, 1500]> var_707_cast_fp16 = slice_by_index(begin = var_707_begin_0, end = var_707_end_0, end_mask = var_707_end_mask_0, x = obj_27_cast_fp16)[name = tensor<string, []>("op_707_cast_fp16")]; |
| 482 | tensor<int32, [4]> var_710_begin_0 = const()[name = tensor<string, []>("op_710_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 483 | tensor<int32, [4]> var_710_end_0 = const()[name = tensor<string, []>("op_710_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| 484 | tensor<bool, [4]> var_710_end_mask_0 = const()[name = tensor<string, []>("op_710_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| 485 | tensor<bool, [4]> var_710_squeeze_mask_0 = const()[name = tensor<string, []>("op_710_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| 486 | tensor<fp16, [1, 1, 1500]> var_710_cast_fp16 = slice_by_index(begin = var_710_begin_0, end = var_710_end_0, end_mask = var_710_end_mask_0, squeeze_mask = var_710_squeeze_mask_0, x = var_707_cast_fp16)[name = tensor<string, []>("op_710_cast_fp16")]; |
| 487 | tensor<int32, [4]> var_725_begin_0 = const()[name = tensor<string, []>("op_725_begin_0"), val = tensor<int32, [4]>([0, 11, 0, 0])]; |
| 488 | tensor<int32, [4]> var_725_end_0 = const()[name = tensor<string, []>("op_725_end_0"), val = tensor<int32, [4]>([1, 12, 1, 1500])]; |
| 489 | tensor<bool, [4]> var_725_end_mask_0 = const()[name = tensor<string, []>("op_725_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| 490 | tensor<fp16, [1, 1, 1, 1500]> var_725_cast_fp16 = slice_by_index(begin = var_725_begin_0, end = var_725_end_0, end_mask = var_725_end_mask_0, x = obj_27_cast_fp16)[name = tensor<string, []>("op_725_cast_fp16")]; |
| 491 | tensor<int32, [4]> var_728_begin_0 = const()[name = tensor<string, []>("op_728_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 492 | tensor<int32, [4]> var_728_end_0 = const()[name = tensor<string, []>("op_728_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| 493 | tensor<bool, [4]> var_728_end_mask_0 = const()[name = tensor<string, []>("op_728_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| 494 | tensor<bool, [4]> var_728_squeeze_mask_0 = const()[name = tensor<string, []>("op_728_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| 495 | tensor<fp16, [1, 1, 1500]> var_728_cast_fp16 = slice_by_index(begin = var_728_begin_0, end = var_728_end_0, end_mask = var_728_end_mask_0, squeeze_mask = var_728_squeeze_mask_0, x = var_725_cast_fp16)[name = tensor<string, []>("op_728_cast_fp16")]; |
| 496 | tensor<int32, [4]> var_743_begin_0 = const()[name = tensor<string, []>("op_743_begin_0"), val = tensor<int32, [4]>([0, 12, 0, 0])]; |
| 497 | tensor<int32, [4]> var_743_end_0 = const()[name = tensor<string, []>("op_743_end_0"), val = tensor<int32, [4]>([1, 13, 1, 1500])]; |
| 498 | tensor<bool, [4]> var_743_end_mask_0 = const()[name = tensor<string, []>("op_743_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| 499 | tensor<fp16, [1, 1, 1, 1500]> var_743_cast_fp16 = slice_by_index(begin = var_743_begin_0, end = var_743_end_0, end_mask = var_743_end_mask_0, x = obj_27_cast_fp16)[name = tensor<string, []>("op_743_cast_fp16")]; |
| 500 | tensor<int32, [4]> var_746_begin_0 = const()[name = tensor<string, []>("op_746_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 501 | tensor<int32, [4]> var_746_end_0 = const()[name = tensor<string, []>("op_746_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| 502 | tensor<bool, [4]> var_746_end_mask_0 = const()[name = tensor<string, []>("op_746_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| 503 | tensor<bool, [4]> var_746_squeeze_mask_0 = const()[name = tensor<string, []>("op_746_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| 504 | tensor<fp16, [1, 1, 1500]> var_746_cast_fp16 = slice_by_index(begin = var_746_begin_0, end = var_746_end_0, end_mask = var_746_end_mask_0, squeeze_mask = var_746_squeeze_mask_0, x = var_743_cast_fp16)[name = tensor<string, []>("op_746_cast_fp16")]; |
| 505 | tensor<int32, [4]> var_761_begin_0 = const()[name = tensor<string, []>("op_761_begin_0"), val = tensor<int32, [4]>([0, 13, 0, 0])]; |
| 506 | tensor<int32, [4]> var_761_end_0 = const()[name = tensor<string, []>("op_761_end_0"), val = tensor<int32, [4]>([1, 14, 1, 1500])]; |
| 507 | tensor<bool, [4]> var_761_end_mask_0 = const()[name = tensor<string, []>("op_761_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| 508 | tensor<fp16, [1, 1, 1, 1500]> var_761_cast_fp16 = slice_by_index(begin = var_761_begin_0, end = var_761_end_0, end_mask = var_761_end_mask_0, x = obj_27_cast_fp16)[name = tensor<string, []>("op_761_cast_fp16")]; |
| 509 | tensor<int32, [4]> var_764_begin_0 = const()[name = tensor<string, []>("op_764_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 510 | tensor<int32, [4]> var_764_end_0 = const()[name = tensor<string, []>("op_764_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| 511 | tensor<bool, [4]> var_764_end_mask_0 = const()[name = tensor<string, []>("op_764_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| 512 | tensor<bool, [4]> var_764_squeeze_mask_0 = const()[name = tensor<string, []>("op_764_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| 513 | tensor<fp16, [1, 1, 1500]> var_764_cast_fp16 = slice_by_index(begin = var_764_begin_0, end = var_764_end_0, end_mask = var_764_end_mask_0, squeeze_mask = var_764_squeeze_mask_0, x = var_761_cast_fp16)[name = tensor<string, []>("op_764_cast_fp16")]; |
| 514 | tensor<int32, [4]> var_779_begin_0 = const()[name = tensor<string, []>("op_779_begin_0"), val = tensor<int32, [4]>([0, 14, 0, 0])]; |
| 515 | tensor<int32, [4]> var_779_end_0 = const()[name = tensor<string, []>("op_779_end_0"), val = tensor<int32, [4]>([1, 15, 1, 1500])]; |
| 516 | tensor<bool, [4]> var_779_end_mask_0 = const()[name = tensor<string, []>("op_779_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| 517 | tensor<fp16, [1, 1, 1, 1500]> var_779_cast_fp16 = slice_by_index(begin = var_779_begin_0, end = var_779_end_0, end_mask = var_779_end_mask_0, x = obj_27_cast_fp16)[name = tensor<string, []>("op_779_cast_fp16")]; |
| 518 | tensor<int32, [4]> var_782_begin_0 = const()[name = tensor<string, []>("op_782_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 519 | tensor<int32, [4]> var_782_end_0 = const()[name = tensor<string, []>("op_782_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| 520 | tensor<bool, [4]> var_782_end_mask_0 = const()[name = tensor<string, []>("op_782_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| 521 | tensor<bool, [4]> var_782_squeeze_mask_0 = const()[name = tensor<string, []>("op_782_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| 522 | tensor<fp16, [1, 1, 1500]> var_782_cast_fp16 = slice_by_index(begin = var_782_begin_0, end = var_782_end_0, end_mask = var_782_end_mask_0, squeeze_mask = var_782_squeeze_mask_0, x = var_779_cast_fp16)[name = tensor<string, []>("op_782_cast_fp16")]; |
| 523 | tensor<int32, [4]> var_797_begin_0 = const()[name = tensor<string, []>("op_797_begin_0"), val = tensor<int32, [4]>([0, 15, 0, 0])]; |
| 524 | tensor<int32, [4]> var_797_end_0 = const()[name = tensor<string, []>("op_797_end_0"), val = tensor<int32, [4]>([1, 16, 1, 1500])]; |
| 525 | tensor<bool, [4]> var_797_end_mask_0 = const()[name = tensor<string, []>("op_797_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| 526 | tensor<fp16, [1, 1, 1, 1500]> var_797_cast_fp16 = slice_by_index(begin = var_797_begin_0, end = var_797_end_0, end_mask = var_797_end_mask_0, x = obj_27_cast_fp16)[name = tensor<string, []>("op_797_cast_fp16")]; |
| 527 | tensor<int32, [4]> var_800_begin_0 = const()[name = tensor<string, []>("op_800_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 528 | tensor<int32, [4]> var_800_end_0 = const()[name = tensor<string, []>("op_800_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| 529 | tensor<bool, [4]> var_800_end_mask_0 = const()[name = tensor<string, []>("op_800_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| 530 | tensor<bool, [4]> var_800_squeeze_mask_0 = const()[name = tensor<string, []>("op_800_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| 531 | tensor<fp16, [1, 1, 1500]> var_800_cast_fp16 = slice_by_index(begin = var_800_begin_0, end = var_800_end_0, end_mask = var_800_end_mask_0, squeeze_mask = var_800_squeeze_mask_0, x = var_797_cast_fp16)[name = tensor<string, []>("op_800_cast_fp16")]; |
| 532 | tensor<int32, [4]> var_815_begin_0 = const()[name = tensor<string, []>("op_815_begin_0"), val = tensor<int32, [4]>([0, 16, 0, 0])]; |
| 533 | tensor<int32, [4]> var_815_end_0 = const()[name = tensor<string, []>("op_815_end_0"), val = tensor<int32, [4]>([1, 17, 1, 1500])]; |
| 534 | tensor<bool, [4]> var_815_end_mask_0 = const()[name = tensor<string, []>("op_815_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| 535 | tensor<fp16, [1, 1, 1, 1500]> var_815_cast_fp16 = slice_by_index(begin = var_815_begin_0, end = var_815_end_0, end_mask = var_815_end_mask_0, x = obj_27_cast_fp16)[name = tensor<string, []>("op_815_cast_fp16")]; |
| 536 | tensor<int32, [4]> var_818_begin_0 = const()[name = tensor<string, []>("op_818_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 537 | tensor<int32, [4]> var_818_end_0 = const()[name = tensor<string, []>("op_818_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| 538 | tensor<bool, [4]> var_818_end_mask_0 = const()[name = tensor<string, []>("op_818_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| 539 | tensor<bool, [4]> var_818_squeeze_mask_0 = const()[name = tensor<string, []>("op_818_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| 540 | tensor<fp16, [1, 1, 1500]> var_818_cast_fp16 = slice_by_index(begin = var_818_begin_0, end = var_818_end_0, end_mask = var_818_end_mask_0, squeeze_mask = var_818_squeeze_mask_0, x = var_815_cast_fp16)[name = tensor<string, []>("op_818_cast_fp16")]; |
| 541 | tensor<int32, [4]> var_833_begin_0 = const()[name = tensor<string, []>("op_833_begin_0"), val = tensor<int32, [4]>([0, 17, 0, 0])]; |
| 542 | tensor<int32, [4]> var_833_end_0 = const()[name = tensor<string, []>("op_833_end_0"), val = tensor<int32, [4]>([1, 18, 1, 1500])]; |
| 543 | tensor<bool, [4]> var_833_end_mask_0 = const()[name = tensor<string, []>("op_833_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| 544 | tensor<fp16, [1, 1, 1, 1500]> var_833_cast_fp16 = slice_by_index(begin = var_833_begin_0, end = var_833_end_0, end_mask = var_833_end_mask_0, x = obj_27_cast_fp16)[name = tensor<string, []>("op_833_cast_fp16")]; |
| 545 | tensor<int32, [4]> var_836_begin_0 = const()[name = tensor<string, []>("op_836_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 546 | tensor<int32, [4]> var_836_end_0 = const()[name = tensor<string, []>("op_836_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| 547 | tensor<bool, [4]> var_836_end_mask_0 = const()[name = tensor<string, []>("op_836_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| 548 | tensor<bool, [4]> var_836_squeeze_mask_0 = const()[name = tensor<string, []>("op_836_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| 549 | tensor<fp16, [1, 1, 1500]> var_836_cast_fp16 = slice_by_index(begin = var_836_begin_0, end = var_836_end_0, end_mask = var_836_end_mask_0, squeeze_mask = var_836_squeeze_mask_0, x = var_833_cast_fp16)[name = tensor<string, []>("op_836_cast_fp16")]; |
| 550 | tensor<int32, [4]> var_851_begin_0 = const()[name = tensor<string, []>("op_851_begin_0"), val = tensor<int32, [4]>([0, 18, 0, 0])]; |
| 551 | tensor<int32, [4]> var_851_end_0 = const()[name = tensor<string, []>("op_851_end_0"), val = tensor<int32, [4]>([1, 19, 1, 1500])]; |
| 552 | tensor<bool, [4]> var_851_end_mask_0 = const()[name = tensor<string, []>("op_851_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| 553 | tensor<fp16, [1, 1, 1, 1500]> var_851_cast_fp16 = slice_by_index(begin = var_851_begin_0, end = var_851_end_0, end_mask = var_851_end_mask_0, x = obj_27_cast_fp16)[name = tensor<string, []>("op_851_cast_fp16")]; |
| 554 | tensor<int32, [4]> var_854_begin_0 = const()[name = tensor<string, []>("op_854_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 555 | tensor<int32, [4]> var_854_end_0 = const()[name = tensor<string, []>("op_854_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| 556 | tensor<bool, [4]> var_854_end_mask_0 = const()[name = tensor<string, []>("op_854_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| 557 | tensor<bool, [4]> var_854_squeeze_mask_0 = const()[name = tensor<string, []>("op_854_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| 558 | tensor<fp16, [1, 1, 1500]> var_854_cast_fp16 = slice_by_index(begin = var_854_begin_0, end = var_854_end_0, end_mask = var_854_end_mask_0, squeeze_mask = var_854_squeeze_mask_0, x = var_851_cast_fp16)[name = tensor<string, []>("op_854_cast_fp16")]; |
| 559 | tensor<int32, [4]> var_869_begin_0 = const()[name = tensor<string, []>("op_869_begin_0"), val = tensor<int32, [4]>([0, 19, 0, 0])]; |
| 560 | tensor<int32, [4]> var_869_end_0 = const()[name = tensor<string, []>("op_869_end_0"), val = tensor<int32, [4]>([1, 20, 1, 1500])]; |
| 561 | tensor<bool, [4]> var_869_end_mask_0 = const()[name = tensor<string, []>("op_869_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])]; |
| 562 | tensor<fp16, [1, 1, 1, 1500]> var_869_cast_fp16 = slice_by_index(begin = var_869_begin_0, end = var_869_end_0, end_mask = var_869_end_mask_0, x = obj_27_cast_fp16)[name = tensor<string, []>("op_869_cast_fp16")]; |
| 563 | tensor<int32, [4]> var_872_begin_0 = const()[name = tensor<string, []>("op_872_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| 564 | tensor<int32, [4]> var_872_end_0 = const()[name = tensor<string, []>("op_872_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1500])]; |
| 565 | tensor<bool, [4]> var_872_end_mask_0 = const()[name = tensor<string, []>("op_872_end_mask_0"), val = tensor<bool, [4]>([true, true, false, true])]; |
| 566 | tensor<bool, [4]> var_872_squeeze_mask_0 = const()[name = tensor<string, []>("op_872_squeeze_mask_0"), val = tensor<bool, [4]>([false, false, true, false])]; |
| 567 | tensor<fp16, [1, 1, 1500]> var_872_cast_fp16 = slice_by_index(begin = var_872_begin_0, end = var_872_end_0, end_mask = var_872_end_mask_0, squeeze_mask = var_872_squeeze_mask_0, x = var_869_cast_fp16)[name = tensor<string, []>("op_872_cast_fp16")]; |
| 568 | tensor<int32, []> var_879 = const()[name = tensor<string, []>("op_879"), val = tensor<int32, []>(1)]; |
| 569 | tensor<bool, []> var_880_interleave_0 = const()[name = tensor<string, []>("op_880_interleave_0"), val = tensor<bool, []>(false)]; |
| 570 | tensor<fp16, [1, 20, 1500]> var_880_cast_fp16 = concat(axis = var_879, interleave = var_880_interleave_0, values = (var_530_cast_fp16, var_548_cast_fp16, var_566_cast_fp16, var_584_cast_fp16, var_602_cast_fp16, var_620_cast_fp16, var_638_cast_fp16, var_656_cast_fp16, var_674_cast_fp16, var_692_cast_fp16, var_710_cast_fp16, var_728_cast_fp16, var_746_cast_fp16, var_764_cast_fp16, var_782_cast_fp16, var_800_cast_fp16, var_818_cast_fp16, var_836_cast_fp16, var_854_cast_fp16, var_872_cast_fp16))[name = tensor<string, []>("op_880_cast_fp16")]; |
| 571 | tensor<int32, [1]> var_882 = const()[name = tensor<string, []>("op_882"), val = tensor<int32, [1]>([1])]; |
| 572 | tensor<bool, []> var_883 = const()[name = tensor<string, []>("op_883"), val = tensor<bool, []>(false)]; |
| 573 | tensor<fp16, [1, 1500]> alignment_heads_weights = reduce_mean(axes = var_882, keep_dims = var_883, x = var_880_cast_fp16)[name = tensor<string, []>("obj_cast_fp16")]; |
| 574 | } -> (logits, key_cache_updates, value_cache_updates, alignment_heads_weights); |
| 575 | } |