{ "architectures": [ "Step5ForConditionalGeneration" ], "model_type": "step5_vl", "im_end_token": "", "im_patch_token": "", "im_start_token": "", "image_token_len": 169, "patch_token_len": 81, "image_token_id": 128001, "understand_projector_stride": 2, "use_im_start_end": "true", "vision_select_layer": -1, "projector_bias": false, "vision_config": { "model_type": "perception_encoder", "image_size": 728, "patch_size": 14, "width": 1536, "layers": 47, "heads": 16, "pool_type": "none", "output_dim": null, "use_cls_token": false, "ls_init_value": 0.1, "use_ln_post": false, "hidden_act": "quick_gelu" }, "text_config": { "architectures": [ "Step5ForCausalLM" ], "model_type": "step5", "att_impl_type": "GQA", "torch_dtype": "bfloat16", "norm_dtype": "float32", "bos_token_id": 0, "eos_token_id": [ 1, 2 ], "hidden_size": 4096, "intermediate_size": 13824, "num_hidden_layers": 5, "vocab_size": 128896, "rms_norm_eps": 1e-05, "fp32_residual_connection": true, "num_attention_heads": 64, "num_attention_groups": 4, "num_sliding_attention_heads": 64, "head_dim": 192, "head_dim_full_attention": 192, "head_dim_sliding_attention": 192, "num_attention_groups_full_attention": 4, "num_attention_groups_sliding_attention": 4, "rope_theta": [ 10000.0, 10000000.0, 10000.0, 10000.0, 10000000.0 ], "layer_types": [ "sliding_attention", "full_attention", "sliding_attention", "sliding_attention", "full_attention" ], "partial_rotary_factors": [ 1.0, 0.3333333333333333, 1.0, 1.0, 0.3333333333333333 ], "sliding_window": 512, "attention_other_setting": { "attention_type": "sliding_attention", "num_attention_heads": 64, "num_attention_groups": 4, "head_dim": 192, "true_head_dim": 192 }, "use_head_wise_attn_gate": true, "num_nextn_predict_layers": 0, "use_moe": true, "moe_intermediate_size": 1536, "share_expert_dim": 1536, "moe_num_experts": 352, "moe_top_k": 8, "num_experts_per_tok": 8, "moe_layer_list": [ 1, 2, 3 ], "moe_layers_enum": "1,2,3", "use_moe_router_bias": true, "need_fp32_gate": true, "swiglu_limits": [ 7.0, 7.0, 7.0, 7.0, 7.0 ], "swiglu_limits_shared": [ 7.0, 7.0, 7.0, 7.0, 7.0 ], "max_seq_len": 1048576, "max_position_embeddings": 1048576, "norm_expert_weight": true, "moe_router_scaling_factor": 3.0, "rope_scaling": { "rope_type": "llama3", "factor": 1.0, "original_max_position_embeddings": 1048576, "low_freq_factor": 1.0, "high_freq_factor": 32.0 }, "yarn_only_types": [ "full_attention" ], "sparse_config": { "enabled": true, "proxy_dim": 256, "sparse_indexer_rope_dim": 32, "sparse_indexer_use_rope": true, "sparse_indexer_num_heads": 16, "sparse_indexer_num_k_heads": 1, "sparse_indexer_q_norm_type": "rmsnorm", "sparse_indexer_k_norm_type": "layernorm", "sparse_indexer_csa_z_norm_type": "none", "sparse_indexer_softmax_variant": "ssmax", "sparse_indexer_ssmax_s_granularity": "q_head", "num_provider_groups": 4, "topk": 512, "region_block_size": 8, "compression_method": "csa_block_compress", "attention_impl": "sparse_gqa", "apply_to_layer_types": [ "full_attention" ] } }, "quantization_config": { "quant_method": "fp8", "activation_scheme": "dynamic", "fmt": "e4m3", "weight_block_size": [ 128, 128 ], "modules_to_not_convert": [ "model.layers.3.moe.experts.w13_weight", "model.layers.3.moe.experts.w2_weight", "vision_model.transformer.resblocks.0.attn.qkv_proj", "vision_model.transformer.resblocks.0.attn.q_proj", "vision_model.transformer.resblocks.0.attn.k_proj", "vision_model.transformer.resblocks.0.attn.v_proj", "vision_model.transformer.resblocks.0.attn.out_proj", "vision_model.transformer.resblocks.0.mlp.fc1", "vision_model.transformer.resblocks.0.mlp.fc2", "vision_model.transformer.resblocks.1.attn.qkv_proj", "vision_model.transformer.resblocks.1.attn.q_proj", "vision_model.transformer.resblocks.1.attn.k_proj", "vision_model.transformer.resblocks.1.attn.v_proj", "vision_model.transformer.resblocks.1.attn.out_proj", "vision_model.transformer.resblocks.1.mlp.fc1", "vision_model.transformer.resblocks.1.mlp.fc2", "vision_model.transformer.resblocks.2.attn.qkv_proj", "vision_model.transformer.resblocks.2.attn.q_proj", "vision_model.transformer.resblocks.2.attn.k_proj", "vision_model.transformer.resblocks.2.attn.v_proj", "vision_model.transformer.resblocks.2.attn.out_proj", "vision_model.transformer.resblocks.2.mlp.fc1", "vision_model.transformer.resblocks.2.mlp.fc2", "vision_model.transformer.resblocks.3.attn.qkv_proj", "vision_model.transformer.resblocks.3.attn.q_proj", "vision_model.transformer.resblocks.3.attn.k_proj", "vision_model.transformer.resblocks.3.attn.v_proj", "vision_model.transformer.resblocks.3.attn.out_proj", "vision_model.transformer.resblocks.3.mlp.fc1", "vision_model.transformer.resblocks.3.mlp.fc2", "vision_model.transformer.resblocks.4.attn.qkv_proj", "vision_model.transformer.resblocks.4.attn.q_proj", "vision_model.transformer.resblocks.4.attn.k_proj", "vision_model.transformer.resblocks.4.attn.v_proj", "vision_model.transformer.resblocks.4.attn.out_proj", "vision_model.transformer.resblocks.4.mlp.fc1", "vision_model.transformer.resblocks.4.mlp.fc2", "vision_model.transformer.resblocks.5.attn.qkv_proj", "vision_model.transformer.resblocks.5.attn.q_proj", "vision_model.transformer.resblocks.5.attn.k_proj", "vision_model.transformer.resblocks.5.attn.v_proj", "vision_model.transformer.resblocks.5.attn.out_proj", "vision_model.transformer.resblocks.5.mlp.fc1", "vision_model.transformer.resblocks.5.mlp.fc2", "vision_model.transformer.resblocks.6.attn.qkv_proj", "vision_model.transformer.resblocks.6.attn.q_proj", "vision_model.transformer.resblocks.6.attn.k_proj", "vision_model.transformer.resblocks.6.attn.v_proj", "vision_model.transformer.resblocks.6.attn.out_proj", "vision_model.transformer.resblocks.6.mlp.fc1", "vision_model.transformer.resblocks.6.mlp.fc2", "vision_model.transformer.resblocks.7.attn.qkv_proj", "vision_model.transformer.resblocks.7.attn.q_proj", "vision_model.transformer.resblocks.7.attn.k_proj", "vision_model.transformer.resblocks.7.attn.v_proj", "vision_model.transformer.resblocks.7.attn.out_proj", "vision_model.transformer.resblocks.7.mlp.fc1", "vision_model.transformer.resblocks.7.mlp.fc2", "vision_model.transformer.resblocks.8.attn.qkv_proj", "vision_model.transformer.resblocks.8.attn.q_proj", "vision_model.transformer.resblocks.8.attn.k_proj", "vision_model.transformer.resblocks.8.attn.v_proj", "vision_model.transformer.resblocks.8.attn.out_proj", "vision_model.transformer.resblocks.8.mlp.fc1", "vision_model.transformer.resblocks.8.mlp.fc2", "vision_model.transformer.resblocks.9.attn.qkv_proj", "vision_model.transformer.resblocks.9.attn.q_proj", "vision_model.transformer.resblocks.9.attn.k_proj", "vision_model.transformer.resblocks.9.attn.v_proj", "vision_model.transformer.resblocks.9.attn.out_proj", "vision_model.transformer.resblocks.9.mlp.fc1", "vision_model.transformer.resblocks.9.mlp.fc2", "vision_model.transformer.resblocks.10.attn.qkv_proj", "vision_model.transformer.resblocks.10.attn.q_proj", "vision_model.transformer.resblocks.10.attn.k_proj", "vision_model.transformer.resblocks.10.attn.v_proj", "vision_model.transformer.resblocks.10.attn.out_proj", "vision_model.transformer.resblocks.10.mlp.fc1", "vision_model.transformer.resblocks.10.mlp.fc2", "vision_model.transformer.resblocks.11.attn.qkv_proj", "vision_model.transformer.resblocks.11.attn.q_proj", "vision_model.transformer.resblocks.11.attn.k_proj", "vision_model.transformer.resblocks.11.attn.v_proj", "vision_model.transformer.resblocks.11.attn.out_proj", "vision_model.transformer.resblocks.11.mlp.fc1", "vision_model.transformer.resblocks.11.mlp.fc2", "vision_model.transformer.resblocks.12.attn.qkv_proj", "vision_model.transformer.resblocks.12.attn.q_proj", "vision_model.transformer.resblocks.12.attn.k_proj", "vision_model.transformer.resblocks.12.attn.v_proj", "vision_model.transformer.resblocks.12.attn.out_proj", "vision_model.transformer.resblocks.12.mlp.fc1", "vision_model.transformer.resblocks.12.mlp.fc2", "vision_model.transformer.resblocks.13.attn.qkv_proj", "vision_model.transformer.resblocks.13.attn.q_proj", "vision_model.transformer.resblocks.13.attn.k_proj", "vision_model.transformer.resblocks.13.attn.v_proj", "vision_model.transformer.resblocks.13.attn.out_proj", "vision_model.transformer.resblocks.13.mlp.fc1", "vision_model.transformer.resblocks.13.mlp.fc2", "vision_model.transformer.resblocks.14.attn.qkv_proj", "vision_model.transformer.resblocks.14.attn.q_proj", "vision_model.transformer.resblocks.14.attn.k_proj", "vision_model.transformer.resblocks.14.attn.v_proj", "vision_model.transformer.resblocks.14.attn.out_proj", "vision_model.transformer.resblocks.14.mlp.fc1", "vision_model.transformer.resblocks.14.mlp.fc2", "vision_model.transformer.resblocks.15.attn.qkv_proj", "vision_model.transformer.resblocks.15.attn.q_proj", "vision_model.transformer.resblocks.15.attn.k_proj", "vision_model.transformer.resblocks.15.attn.v_proj", "vision_model.transformer.resblocks.15.attn.out_proj", "vision_model.transformer.resblocks.15.mlp.fc1", "vision_model.transformer.resblocks.15.mlp.fc2", "vision_model.transformer.resblocks.16.attn.qkv_proj", "vision_model.transformer.resblocks.16.attn.q_proj", "vision_model.transformer.resblocks.16.attn.k_proj", "vision_model.transformer.resblocks.16.attn.v_proj", "vision_model.transformer.resblocks.16.attn.out_proj", "vision_model.transformer.resblocks.16.mlp.fc1", "vision_model.transformer.resblocks.16.mlp.fc2", "vision_model.transformer.resblocks.17.attn.qkv_proj", "vision_model.transformer.resblocks.17.attn.q_proj", "vision_model.transformer.resblocks.17.attn.k_proj", "vision_model.transformer.resblocks.17.attn.v_proj", "vision_model.transformer.resblocks.17.attn.out_proj", "vision_model.transformer.resblocks.17.mlp.fc1", "vision_model.transformer.resblocks.17.mlp.fc2", "vision_model.transformer.resblocks.18.attn.qkv_proj", "vision_model.transformer.resblocks.18.attn.q_proj", "vision_model.transformer.resblocks.18.attn.k_proj", "vision_model.transformer.resblocks.18.attn.v_proj", "vision_model.transformer.resblocks.18.attn.out_proj", "vision_model.transformer.resblocks.18.mlp.fc1", "vision_model.transformer.resblocks.18.mlp.fc2", "vision_model.transformer.resblocks.19.attn.qkv_proj", "vision_model.transformer.resblocks.19.attn.q_proj", "vision_model.transformer.resblocks.19.attn.k_proj", "vision_model.transformer.resblocks.19.attn.v_proj", "vision_model.transformer.resblocks.19.attn.out_proj", "vision_model.transformer.resblocks.19.mlp.fc1", "vision_model.transformer.resblocks.19.mlp.fc2", "vision_model.transformer.resblocks.20.attn.qkv_proj", "vision_model.transformer.resblocks.20.attn.q_proj", "vision_model.transformer.resblocks.20.attn.k_proj", "vision_model.transformer.resblocks.20.attn.v_proj", "vision_model.transformer.resblocks.20.attn.out_proj", "vision_model.transformer.resblocks.20.mlp.fc1", "vision_model.transformer.resblocks.20.mlp.fc2", "vision_model.transformer.resblocks.21.attn.qkv_proj", "vision_model.transformer.resblocks.21.attn.q_proj", "vision_model.transformer.resblocks.21.attn.k_proj", "vision_model.transformer.resblocks.21.attn.v_proj", "vision_model.transformer.resblocks.21.attn.out_proj", "vision_model.transformer.resblocks.21.mlp.fc1", "vision_model.transformer.resblocks.21.mlp.fc2", "vision_model.transformer.resblocks.22.attn.qkv_proj", "vision_model.transformer.resblocks.22.attn.q_proj", "vision_model.transformer.resblocks.22.attn.k_proj", "vision_model.transformer.resblocks.22.attn.v_proj", "vision_model.transformer.resblocks.22.attn.out_proj", "vision_model.transformer.resblocks.22.mlp.fc1", "vision_model.transformer.resblocks.22.mlp.fc2", "vision_model.transformer.resblocks.23.attn.qkv_proj", "vision_model.transformer.resblocks.23.attn.q_proj", "vision_model.transformer.resblocks.23.attn.k_proj", "vision_model.transformer.resblocks.23.attn.v_proj", "vision_model.transformer.resblocks.23.attn.out_proj", "vision_model.transformer.resblocks.23.mlp.fc1", "vision_model.transformer.resblocks.23.mlp.fc2", "vision_model.transformer.resblocks.24.attn.qkv_proj", "vision_model.transformer.resblocks.24.attn.q_proj", "vision_model.transformer.resblocks.24.attn.k_proj", "vision_model.transformer.resblocks.24.attn.v_proj", "vision_model.transformer.resblocks.24.attn.out_proj", "vision_model.transformer.resblocks.24.mlp.fc1", "vision_model.transformer.resblocks.24.mlp.fc2", "vision_model.transformer.resblocks.25.attn.qkv_proj", "vision_model.transformer.resblocks.25.attn.q_proj", "vision_model.transformer.resblocks.25.attn.k_proj", "vision_model.transformer.resblocks.25.attn.v_proj", "vision_model.transformer.resblocks.25.attn.out_proj", "vision_model.transformer.resblocks.25.mlp.fc1", "vision_model.transformer.resblocks.25.mlp.fc2", "vision_model.transformer.resblocks.26.attn.qkv_proj", "vision_model.transformer.resblocks.26.attn.q_proj", "vision_model.transformer.resblocks.26.attn.k_proj", "vision_model.transformer.resblocks.26.attn.v_proj", "vision_model.transformer.resblocks.26.attn.out_proj", "vision_model.transformer.resblocks.26.mlp.fc1", "vision_model.transformer.resblocks.26.mlp.fc2", "vision_model.transformer.resblocks.27.attn.qkv_proj", "vision_model.transformer.resblocks.27.attn.q_proj", "vision_model.transformer.resblocks.27.attn.k_proj", "vision_model.transformer.resblocks.27.attn.v_proj", "vision_model.transformer.resblocks.27.attn.out_proj", "vision_model.transformer.resblocks.27.mlp.fc1", "vision_model.transformer.resblocks.27.mlp.fc2", "vision_model.transformer.resblocks.28.attn.qkv_proj", "vision_model.transformer.resblocks.28.attn.q_proj", "vision_model.transformer.resblocks.28.attn.k_proj", "vision_model.transformer.resblocks.28.attn.v_proj", "vision_model.transformer.resblocks.28.attn.out_proj", "vision_model.transformer.resblocks.28.mlp.fc1", "vision_model.transformer.resblocks.28.mlp.fc2", "vision_model.transformer.resblocks.29.attn.qkv_proj", "vision_model.transformer.resblocks.29.attn.q_proj", "vision_model.transformer.resblocks.29.attn.k_proj", "vision_model.transformer.resblocks.29.attn.v_proj", "vision_model.transformer.resblocks.29.attn.out_proj", "vision_model.transformer.resblocks.29.mlp.fc1", "vision_model.transformer.resblocks.29.mlp.fc2", "vision_model.transformer.resblocks.30.attn.qkv_proj", "vision_model.transformer.resblocks.30.attn.q_proj", "vision_model.transformer.resblocks.30.attn.k_proj", "vision_model.transformer.resblocks.30.attn.v_proj", "vision_model.transformer.resblocks.30.attn.out_proj", "vision_model.transformer.resblocks.30.mlp.fc1", "vision_model.transformer.resblocks.30.mlp.fc2", "vision_model.transformer.resblocks.31.attn.qkv_proj", "vision_model.transformer.resblocks.31.attn.q_proj", "vision_model.transformer.resblocks.31.attn.k_proj", "vision_model.transformer.resblocks.31.attn.v_proj", "vision_model.transformer.resblocks.31.attn.out_proj", "vision_model.transformer.resblocks.31.mlp.fc1", "vision_model.transformer.resblocks.31.mlp.fc2", "vision_model.transformer.resblocks.32.attn.qkv_proj", "vision_model.transformer.resblocks.32.attn.q_proj", "vision_model.transformer.resblocks.32.attn.k_proj", "vision_model.transformer.resblocks.32.attn.v_proj", "vision_model.transformer.resblocks.32.attn.out_proj", "vision_model.transformer.resblocks.32.mlp.fc1", "vision_model.transformer.resblocks.32.mlp.fc2", "vision_model.transformer.resblocks.33.attn.qkv_proj", "vision_model.transformer.resblocks.33.attn.q_proj", "vision_model.transformer.resblocks.33.attn.k_proj", "vision_model.transformer.resblocks.33.attn.v_proj", "vision_model.transformer.resblocks.33.attn.out_proj", "vision_model.transformer.resblocks.33.mlp.fc1", "vision_model.transformer.resblocks.33.mlp.fc2", "vision_model.transformer.resblocks.34.attn.qkv_proj", "vision_model.transformer.resblocks.34.attn.q_proj", "vision_model.transformer.resblocks.34.attn.k_proj", "vision_model.transformer.resblocks.34.attn.v_proj", "vision_model.transformer.resblocks.34.attn.out_proj", "vision_model.transformer.resblocks.34.mlp.fc1", "vision_model.transformer.resblocks.34.mlp.fc2", "vision_model.transformer.resblocks.35.attn.qkv_proj", "vision_model.transformer.resblocks.35.attn.q_proj", "vision_model.transformer.resblocks.35.attn.k_proj", "vision_model.transformer.resblocks.35.attn.v_proj", "vision_model.transformer.resblocks.35.attn.out_proj", "vision_model.transformer.resblocks.35.mlp.fc1", "vision_model.transformer.resblocks.35.mlp.fc2", "vision_model.transformer.resblocks.36.attn.qkv_proj", "vision_model.transformer.resblocks.36.attn.q_proj", "vision_model.transformer.resblocks.36.attn.k_proj", "vision_model.transformer.resblocks.36.attn.v_proj", "vision_model.transformer.resblocks.36.attn.out_proj", "vision_model.transformer.resblocks.36.mlp.fc1", "vision_model.transformer.resblocks.36.mlp.fc2", "vision_model.transformer.resblocks.37.attn.qkv_proj", "vision_model.transformer.resblocks.37.attn.q_proj", "vision_model.transformer.resblocks.37.attn.k_proj", "vision_model.transformer.resblocks.37.attn.v_proj", "vision_model.transformer.resblocks.37.attn.out_proj", "vision_model.transformer.resblocks.37.mlp.fc1", "vision_model.transformer.resblocks.37.mlp.fc2", "vision_model.transformer.resblocks.38.attn.qkv_proj", "vision_model.transformer.resblocks.38.attn.q_proj", "vision_model.transformer.resblocks.38.attn.k_proj", "vision_model.transformer.resblocks.38.attn.v_proj", "vision_model.transformer.resblocks.38.attn.out_proj", "vision_model.transformer.resblocks.38.mlp.fc1", "vision_model.transformer.resblocks.38.mlp.fc2", "vision_model.transformer.resblocks.39.attn.qkv_proj", "vision_model.transformer.resblocks.39.attn.q_proj", "vision_model.transformer.resblocks.39.attn.k_proj", "vision_model.transformer.resblocks.39.attn.v_proj", "vision_model.transformer.resblocks.39.attn.out_proj", "vision_model.transformer.resblocks.39.mlp.fc1", "vision_model.transformer.resblocks.39.mlp.fc2", "vision_model.transformer.resblocks.40.attn.qkv_proj", "vision_model.transformer.resblocks.40.attn.q_proj", "vision_model.transformer.resblocks.40.attn.k_proj", "vision_model.transformer.resblocks.40.attn.v_proj", "vision_model.transformer.resblocks.40.attn.out_proj", "vision_model.transformer.resblocks.40.mlp.fc1", "vision_model.transformer.resblocks.40.mlp.fc2", "vision_model.transformer.resblocks.41.attn.qkv_proj", "vision_model.transformer.resblocks.41.attn.q_proj", "vision_model.transformer.resblocks.41.attn.k_proj", "vision_model.transformer.resblocks.41.attn.v_proj", "vision_model.transformer.resblocks.41.attn.out_proj", "vision_model.transformer.resblocks.41.mlp.fc1", "vision_model.transformer.resblocks.41.mlp.fc2", "vision_model.transformer.resblocks.42.attn.qkv_proj", "vision_model.transformer.resblocks.42.attn.q_proj", "vision_model.transformer.resblocks.42.attn.k_proj", "vision_model.transformer.resblocks.42.attn.v_proj", "vision_model.transformer.resblocks.42.attn.out_proj", "vision_model.transformer.resblocks.42.mlp.fc1", "vision_model.transformer.resblocks.42.mlp.fc2", "vision_model.transformer.resblocks.43.attn.qkv_proj", "vision_model.transformer.resblocks.43.attn.q_proj", "vision_model.transformer.resblocks.43.attn.k_proj", "vision_model.transformer.resblocks.43.attn.v_proj", "vision_model.transformer.resblocks.43.attn.out_proj", "vision_model.transformer.resblocks.43.mlp.fc1", "vision_model.transformer.resblocks.43.mlp.fc2", "vision_model.transformer.resblocks.44.attn.qkv_proj", "vision_model.transformer.resblocks.44.attn.q_proj", "vision_model.transformer.resblocks.44.attn.k_proj", "vision_model.transformer.resblocks.44.attn.v_proj", "vision_model.transformer.resblocks.44.attn.out_proj", "vision_model.transformer.resblocks.44.mlp.fc1", "vision_model.transformer.resblocks.44.mlp.fc2", "vision_model.transformer.resblocks.45.attn.qkv_proj", "vision_model.transformer.resblocks.45.attn.q_proj", "vision_model.transformer.resblocks.45.attn.k_proj", "vision_model.transformer.resblocks.45.attn.v_proj", "vision_model.transformer.resblocks.45.attn.out_proj", "vision_model.transformer.resblocks.45.mlp.fc1", "vision_model.transformer.resblocks.45.mlp.fc2", "vision_model.transformer.resblocks.46.attn.qkv_proj", "vision_model.transformer.resblocks.46.attn.q_proj", "vision_model.transformer.resblocks.46.attn.k_proj", "vision_model.transformer.resblocks.46.attn.v_proj", "vision_model.transformer.resblocks.46.attn.out_proj", "vision_model.transformer.resblocks.46.mlp.fc1", "vision_model.transformer.resblocks.46.mlp.fc2", "vit_large_projector", "lm_head", "model.embed_tokens", "model.norm", "model.layers.0.self_attn.g_proj", "model.layers.0.self_attn.qkv_proj", "model.layers.0.self_attn.q_proj", "model.layers.0.self_attn.k_proj", "model.layers.0.self_attn.v_proj", "model.layers.0.self_attn.o_proj", "model.layers.0.mlp.gate_up_proj", "model.layers.0.mlp.gate_proj", "model.layers.0.mlp.up_proj", "model.layers.0.mlp.down_proj", "model.layers.1.self_attn.g_proj", "model.layers.1.self_attn.qkv_proj", "model.layers.1.self_attn.q_proj", "model.layers.1.self_attn.k_proj", "model.layers.1.self_attn.v_proj", "model.layers.1.self_attn.o_proj", "model.layers.1.self_attn.qkv_indexer_proj", "model.layers.1.moe.gate", "model.layers.1.share_expert.gate_up_proj", "model.layers.1.share_expert.gate_proj", "model.layers.1.share_expert.up_proj", "model.layers.1.share_expert.down_proj", "model.layers.2.self_attn.g_proj", "model.layers.2.self_attn.qkv_proj", "model.layers.2.self_attn.q_proj", "model.layers.2.self_attn.k_proj", "model.layers.2.self_attn.v_proj", "model.layers.2.self_attn.o_proj", "model.layers.2.moe.gate", "model.layers.2.share_expert.gate_up_proj", "model.layers.2.share_expert.gate_proj", "model.layers.2.share_expert.up_proj", "model.layers.2.share_expert.down_proj", "model.layers.3.self_attn.g_proj", "model.layers.3.self_attn.qkv_proj", "model.layers.3.self_attn.q_proj", "model.layers.3.self_attn.k_proj", "model.layers.3.self_attn.v_proj", "model.layers.3.self_attn.o_proj", "model.layers.3.moe.gate", "model.layers.3.share_expert.gate_up_proj", "model.layers.3.share_expert.gate_proj", "model.layers.3.share_expert.up_proj", "model.layers.3.share_expert.down_proj", "model.layers.4.self_attn.g_proj", "model.layers.4.self_attn.qkv_proj", "model.layers.4.self_attn.q_proj", "model.layers.4.self_attn.k_proj", "model.layers.4.self_attn.v_proj", "model.layers.4.self_attn.o_proj", "model.layers.4.self_attn.qkv_indexer_proj", "model.layers.4.mlp.gate_up_proj", "model.layers.4.mlp.gate_proj", "model.layers.4.mlp.up_proj", "model.layers.4.mlp.down_proj" ] } }