| { | |
| "architectures": [ | |
| "KeyeVLMoeForConditionalGeneration" | |
| ], | |
| "attention_bias": false, | |
| "attention_dropout": 0.0, | |
| "auto_map": { | |
| "AutoConfig": "configuration_deepseek.KeyeVLMoeConfig" | |
| }, | |
| "bos_token_id": 0, | |
| "dtype": "bfloat16", | |
| "eos_token_id": 1, | |
| "ep_size": 1, | |
| "fast_video_token_id": 128021, | |
| "first_k_dense_replace": 3, | |
| "head_dim": 128, | |
| "hidden_act": "silu", | |
| "hidden_size": 7168, | |
| "image_token_id": 128010, | |
| "initializer_range": 0.02, | |
| "intermediate_size": 18432, | |
| "kv_lora_rank": 512, | |
| "max_position_embeddings": 163840, | |
| "model_type": "KeyeVL", | |
| "moe_intermediate_size": 2048, | |
| "moe_layer_freq": 1, | |
| "n_group": 8, | |
| "n_routed_experts": 256, | |
| "n_shared_experts": 1, | |
| "norm_topk_prob": true, | |
| "num_attention_heads": 128, | |
| "num_experts_per_tok": 8, | |
| "num_hidden_layers": 61, | |
| "num_key_value_heads": 128, | |
| "num_nextn_predict_layers": 1, | |
| "q_lora_rank": 1536, | |
| "qk_nope_head_dim": 128, | |
| "qk_rope_head_dim": 64, | |
| "quantization_config": { | |
| "activation_scheme": "dynamic", | |
| "fmt": "e4m3", | |
| "quant_method": "fp8", | |
| "ignored_layers": [ | |
| "mlp_AR.linear_1", | |
| "mlp_AR.linear_2", | |
| "visual.vision_model.encoder.layers.0.self_attn.k_proj", | |
| "visual.vision_model.encoder.layers.0.self_attn.v_proj", | |
| "visual.vision_model.encoder.layers.0.self_attn.q_proj", | |
| "visual.vision_model.encoder.layers.0.self_attn.out_proj", | |
| "visual.vision_model.encoder.layers.0.mlp.fc1", | |
| "visual.vision_model.encoder.layers.0.mlp.fc2", | |
| "visual.vision_model.encoder.layers.1.self_attn.k_proj", | |
| "visual.vision_model.encoder.layers.1.self_attn.v_proj", | |
| "visual.vision_model.encoder.layers.1.self_attn.q_proj", | |
| "visual.vision_model.encoder.layers.1.self_attn.out_proj", | |
| "visual.vision_model.encoder.layers.1.mlp.fc1", | |
| "visual.vision_model.encoder.layers.1.mlp.fc2", | |
| "visual.vision_model.encoder.layers.2.self_attn.k_proj", | |
| "visual.vision_model.encoder.layers.2.self_attn.v_proj", | |
| "visual.vision_model.encoder.layers.2.self_attn.q_proj", | |
| "visual.vision_model.encoder.layers.2.self_attn.out_proj", | |
| "visual.vision_model.encoder.layers.2.mlp.fc1", | |
| "visual.vision_model.encoder.layers.2.mlp.fc2", | |
| "visual.vision_model.encoder.layers.3.self_attn.k_proj", | |
| "visual.vision_model.encoder.layers.3.self_attn.v_proj", | |
| "visual.vision_model.encoder.layers.3.self_attn.q_proj", | |
| "visual.vision_model.encoder.layers.3.self_attn.out_proj", | |
| "visual.vision_model.encoder.layers.3.mlp.fc1", | |
| "visual.vision_model.encoder.layers.3.mlp.fc2", | |
| "visual.vision_model.encoder.layers.4.self_attn.k_proj", | |
| "visual.vision_model.encoder.layers.4.self_attn.v_proj", | |
| "visual.vision_model.encoder.layers.4.self_attn.q_proj", | |
| "visual.vision_model.encoder.layers.4.self_attn.out_proj", | |
| "visual.vision_model.encoder.layers.4.mlp.fc1", | |
| "visual.vision_model.encoder.layers.4.mlp.fc2", | |
| "visual.vision_model.encoder.layers.5.self_attn.k_proj", | |
| "visual.vision_model.encoder.layers.5.self_attn.v_proj", | |
| "visual.vision_model.encoder.layers.5.self_attn.q_proj", | |
| "visual.vision_model.encoder.layers.5.self_attn.out_proj", | |
| "visual.vision_model.encoder.layers.5.mlp.fc1", | |
| "visual.vision_model.encoder.layers.5.mlp.fc2", | |
| "visual.vision_model.encoder.layers.6.self_attn.k_proj", | |
| "visual.vision_model.encoder.layers.6.self_attn.v_proj", | |
| "visual.vision_model.encoder.layers.6.self_attn.q_proj", | |
| "visual.vision_model.encoder.layers.6.self_attn.out_proj", | |
| "visual.vision_model.encoder.layers.6.mlp.fc1", | |
| "visual.vision_model.encoder.layers.6.mlp.fc2", | |
| "visual.vision_model.encoder.layers.7.self_attn.k_proj", | |
| "visual.vision_model.encoder.layers.7.self_attn.v_proj", | |
| "visual.vision_model.encoder.layers.7.self_attn.q_proj", | |
| "visual.vision_model.encoder.layers.7.self_attn.out_proj", | |
| "visual.vision_model.encoder.layers.7.mlp.fc1", | |
| "visual.vision_model.encoder.layers.7.mlp.fc2", | |
| "visual.vision_model.encoder.layers.8.self_attn.k_proj", | |
| "visual.vision_model.encoder.layers.8.self_attn.v_proj", | |
| "visual.vision_model.encoder.layers.8.self_attn.q_proj", | |
| "visual.vision_model.encoder.layers.8.self_attn.out_proj", | |
| "visual.vision_model.encoder.layers.8.mlp.fc1", | |
| "visual.vision_model.encoder.layers.8.mlp.fc2", | |
| "visual.vision_model.encoder.layers.9.self_attn.k_proj", | |
| "visual.vision_model.encoder.layers.9.self_attn.v_proj", | |
| "visual.vision_model.encoder.layers.9.self_attn.q_proj", | |
| "visual.vision_model.encoder.layers.9.self_attn.out_proj", | |
| "visual.vision_model.encoder.layers.9.mlp.fc1", | |
| "visual.vision_model.encoder.layers.9.mlp.fc2", | |
| "visual.vision_model.encoder.layers.10.self_attn.k_proj", | |
| "visual.vision_model.encoder.layers.10.self_attn.v_proj", | |
| "visual.vision_model.encoder.layers.10.self_attn.q_proj", | |
| "visual.vision_model.encoder.layers.10.self_attn.out_proj", | |
| "visual.vision_model.encoder.layers.10.mlp.fc1", | |
| "visual.vision_model.encoder.layers.10.mlp.fc2", | |
| "visual.vision_model.encoder.layers.11.self_attn.k_proj", | |
| "visual.vision_model.encoder.layers.11.self_attn.v_proj", | |
| "visual.vision_model.encoder.layers.11.self_attn.q_proj", | |
| "visual.vision_model.encoder.layers.11.self_attn.out_proj", | |
| "visual.vision_model.encoder.layers.11.mlp.fc1", | |
| "visual.vision_model.encoder.layers.11.mlp.fc2", | |
| "visual.vision_model.encoder.layers.12.self_attn.k_proj", | |
| "visual.vision_model.encoder.layers.12.self_attn.v_proj", | |
| "visual.vision_model.encoder.layers.12.self_attn.q_proj", | |
| "visual.vision_model.encoder.layers.12.self_attn.out_proj", | |
| "visual.vision_model.encoder.layers.12.mlp.fc1", | |
| "visual.vision_model.encoder.layers.12.mlp.fc2", | |
| "visual.vision_model.encoder.layers.13.self_attn.k_proj", | |
| "visual.vision_model.encoder.layers.13.self_attn.v_proj", | |
| "visual.vision_model.encoder.layers.13.self_attn.q_proj", | |
| "visual.vision_model.encoder.layers.13.self_attn.out_proj", | |
| "visual.vision_model.encoder.layers.13.mlp.fc1", | |
| "visual.vision_model.encoder.layers.13.mlp.fc2", | |
| "visual.vision_model.encoder.layers.14.self_attn.k_proj", | |
| "visual.vision_model.encoder.layers.14.self_attn.v_proj", | |
| "visual.vision_model.encoder.layers.14.self_attn.q_proj", | |
| "visual.vision_model.encoder.layers.14.self_attn.out_proj", | |
| "visual.vision_model.encoder.layers.14.mlp.fc1", | |
| "visual.vision_model.encoder.layers.14.mlp.fc2", | |
| "visual.vision_model.encoder.layers.15.self_attn.k_proj", | |
| "visual.vision_model.encoder.layers.15.self_attn.v_proj", | |
| "visual.vision_model.encoder.layers.15.self_attn.q_proj", | |
| "visual.vision_model.encoder.layers.15.self_attn.out_proj", | |
| "visual.vision_model.encoder.layers.15.mlp.fc1", | |
| "visual.vision_model.encoder.layers.15.mlp.fc2", | |
| "visual.vision_model.encoder.layers.16.self_attn.k_proj", | |
| "visual.vision_model.encoder.layers.16.self_attn.v_proj", | |
| "visual.vision_model.encoder.layers.16.self_attn.q_proj", | |
| "visual.vision_model.encoder.layers.16.self_attn.out_proj", | |
| "visual.vision_model.encoder.layers.16.mlp.fc1", | |
| "visual.vision_model.encoder.layers.16.mlp.fc2", | |
| "visual.vision_model.encoder.layers.17.self_attn.k_proj", | |
| "visual.vision_model.encoder.layers.17.self_attn.v_proj", | |
| "visual.vision_model.encoder.layers.17.self_attn.q_proj", | |
| "visual.vision_model.encoder.layers.17.self_attn.out_proj", | |
| "visual.vision_model.encoder.layers.17.mlp.fc1", | |
| "visual.vision_model.encoder.layers.17.mlp.fc2", | |
| "visual.vision_model.encoder.layers.18.self_attn.k_proj", | |
| "visual.vision_model.encoder.layers.18.self_attn.v_proj", | |
| "visual.vision_model.encoder.layers.18.self_attn.q_proj", | |
| "visual.vision_model.encoder.layers.18.self_attn.out_proj", | |
| "visual.vision_model.encoder.layers.18.mlp.fc1", | |
| "visual.vision_model.encoder.layers.18.mlp.fc2", | |
| "visual.vision_model.encoder.layers.19.self_attn.k_proj", | |
| "visual.vision_model.encoder.layers.19.self_attn.v_proj", | |
| "visual.vision_model.encoder.layers.19.self_attn.q_proj", | |
| "visual.vision_model.encoder.layers.19.self_attn.out_proj", | |
| "visual.vision_model.encoder.layers.19.mlp.fc1", | |
| "visual.vision_model.encoder.layers.19.mlp.fc2", | |
| "visual.vision_model.encoder.layers.20.self_attn.k_proj", | |
| "visual.vision_model.encoder.layers.20.self_attn.v_proj", | |
| "visual.vision_model.encoder.layers.20.self_attn.q_proj", | |
| "visual.vision_model.encoder.layers.20.self_attn.out_proj", | |
| "visual.vision_model.encoder.layers.20.mlp.fc1", | |
| "visual.vision_model.encoder.layers.20.mlp.fc2", | |
| "visual.vision_model.encoder.layers.21.self_attn.k_proj", | |
| "visual.vision_model.encoder.layers.21.self_attn.v_proj", | |
| "visual.vision_model.encoder.layers.21.self_attn.q_proj", | |
| "visual.vision_model.encoder.layers.21.self_attn.out_proj", | |
| "visual.vision_model.encoder.layers.21.mlp.fc1", | |
| "visual.vision_model.encoder.layers.21.mlp.fc2", | |
| "visual.vision_model.encoder.layers.22.self_attn.k_proj", | |
| "visual.vision_model.encoder.layers.22.self_attn.v_proj", | |
| "visual.vision_model.encoder.layers.22.self_attn.q_proj", | |
| "visual.vision_model.encoder.layers.22.self_attn.out_proj", | |
| "visual.vision_model.encoder.layers.22.mlp.fc1", | |
| "visual.vision_model.encoder.layers.22.mlp.fc2", | |
| "visual.vision_model.encoder.layers.23.self_attn.k_proj", | |
| "visual.vision_model.encoder.layers.23.self_attn.v_proj", | |
| "visual.vision_model.encoder.layers.23.self_attn.q_proj", | |
| "visual.vision_model.encoder.layers.23.self_attn.out_proj", | |
| "visual.vision_model.encoder.layers.23.mlp.fc1", | |
| "visual.vision_model.encoder.layers.23.mlp.fc2", | |
| "visual.vision_model.encoder.layers.24.self_attn.k_proj", | |
| "visual.vision_model.encoder.layers.24.self_attn.v_proj", | |
| "visual.vision_model.encoder.layers.24.self_attn.q_proj", | |
| "visual.vision_model.encoder.layers.24.self_attn.out_proj", | |
| "visual.vision_model.encoder.layers.24.mlp.fc1", | |
| "visual.vision_model.encoder.layers.24.mlp.fc2", | |
| "visual.vision_model.encoder.layers.25.self_attn.k_proj", | |
| "visual.vision_model.encoder.layers.25.self_attn.v_proj", | |
| "visual.vision_model.encoder.layers.25.self_attn.q_proj", | |
| "visual.vision_model.encoder.layers.25.self_attn.out_proj", | |
| "visual.vision_model.encoder.layers.25.mlp.fc1", | |
| "visual.vision_model.encoder.layers.25.mlp.fc2", | |
| "visual.vision_model.encoder.layers.26.self_attn.k_proj", | |
| "visual.vision_model.encoder.layers.26.self_attn.v_proj", | |
| "visual.vision_model.encoder.layers.26.self_attn.q_proj", | |
| "visual.vision_model.encoder.layers.26.self_attn.out_proj", | |
| "visual.vision_model.encoder.layers.26.mlp.fc1", | |
| "visual.vision_model.encoder.layers.26.mlp.fc2", | |
| "lm_head" | |
| ], | |
| "weight_block_size": [ | |
| 128, | |
| 128 | |
| ], | |
| "modules_to_not_convert": [ | |
| "mlp_AR", | |
| "visual.vision_model", | |
| "lm_head" | |
| ] | |
| }, | |
| "rms_norm_eps": 1e-06, | |
| "rope_scaling": { | |
| "beta_fast": 32, | |
| "beta_slow": 1, | |
| "factor": 40, | |
| "mscale": 1.0, | |
| "mscale_all_dim": 1.0, | |
| "original_max_position_embeddings": 4096, | |
| "type": "yarn" | |
| }, | |
| "rope_theta": 10000, | |
| "routed_scaling_factor": 2.5, | |
| "scoring_func": "sigmoid", | |
| "tie_word_embeddings": false, | |
| "topk_group": 4, | |
| "topk_method": "noaux_tc", | |
| "transformers_version": "4.56.2", | |
| "use_cache": true, | |
| "v_head_dim": 128, | |
| "video_token_id": 128011, | |
| "vision_config": { | |
| "_attn_implementation_autoset": true, | |
| "architectures": [ | |
| "SiglipVisionModel" | |
| ], | |
| "attention_dropout": 0.0, | |
| "auto_map": { | |
| "AutoConfig": "configuration_deepseek.KeyeVisionConfig" | |
| }, | |
| "has_learnable_position_embedding": true, | |
| "hidden_act": "gelu_pytorch_tanh", | |
| "hidden_size": 1152, | |
| "image_size": 384, | |
| "intermediate_size": 4304, | |
| "layer_norm_eps": 1e-06, | |
| "model_type": "Keye", | |
| "num_attention_heads": 16, | |
| "num_channels": 3, | |
| "num_hidden_layers": 27, | |
| "patch_size": 14, | |
| "rope_theta": 10000, | |
| "spatial_merge_size": 2, | |
| "temporal_patch_size": 2, | |
| "tokens_per_second": 2 | |
| }, | |
| "vision_end_token_id": 128008, | |
| "vision_start_token_id": 128007, | |
| "vision_token_id": 128009, | |
| "vocab_size": 129280 | |
| } | |