config.json
| 1 | { |
| 2 | "architectures": [ |
| 3 | "GlmOcrForConditionalGeneration" |
| 4 | ], |
| 5 | "model_type": "glm_ocr", |
| 6 | "text_config": { |
| 7 | "model_type": "glm_ocr_text", |
| 8 | "pad_token_id": 59246, |
| 9 | "vocab_size": 59392, |
| 10 | "eos_token_id": [ |
| 11 | 59246, |
| 12 | 59253 |
| 13 | ], |
| 14 | "attention_bias": false, |
| 15 | "attention_dropout": 0.0, |
| 16 | "head_dim": 128, |
| 17 | "hidden_act": "silu", |
| 18 | "hidden_size": 1536, |
| 19 | "initializer_range": 0.02, |
| 20 | "intermediate_size": 4608, |
| 21 | "max_position_embeddings": 131072, |
| 22 | "num_attention_heads": 16, |
| 23 | "num_hidden_layers": 16, |
| 24 | "num_nextn_predict_layers": 1, |
| 25 | "num_key_value_heads": 8, |
| 26 | "rms_norm_eps": 1e-05, |
| 27 | "dtype": "bfloat16", |
| 28 | "rope_parameters": { |
| 29 | "rope_type": "default", |
| 30 | "mrope_section": [ |
| 31 | 16, |
| 32 | 24, |
| 33 | 24 |
| 34 | ], |
| 35 | "partial_rotary_factor": 1.0, |
| 36 | "rope_theta": 10000 |
| 37 | }, |
| 38 | "tie_word_embeddings": false, |
| 39 | "use_cache": true |
| 40 | }, |
| 41 | "vision_config": { |
| 42 | "model_type": "glm_ocr_vision", |
| 43 | "hidden_size": 1024, |
| 44 | "depth": 24, |
| 45 | "num_heads": 16, |
| 46 | "attention_bias": true, |
| 47 | "intermediate_size": 4096, |
| 48 | "hidden_act": "silu", |
| 49 | "hidden_dropout_prob": 0.0, |
| 50 | "initializer_range": 0.02, |
| 51 | "image_size": 336, |
| 52 | "patch_size": 14, |
| 53 | "out_hidden_size": 1536, |
| 54 | "rms_norm_eps": 1e-05, |
| 55 | "spatial_merge_size": 2, |
| 56 | "temporal_patch_size": 2 |
| 57 | }, |
| 58 | "image_start_token_id": 59256, |
| 59 | "image_end_token_id": 59257, |
| 60 | "video_start_token_id": 59258, |
| 61 | "video_end_token_id": 59259, |
| 62 | "image_token_id": 59280, |
| 63 | "video_token_id": 59281, |
| 64 | "transformers_version": "5.0.1dev0" |
| 65 | } |
| 66 | |