{ "architectures": [ "MiniMaxM3SparseForConditionalGeneration" ], "auto_map": { "AutoConfig": "configuration_minimax_m3_vl.MiniMaxM3VLConfig" }, "model_type": "minimax_m3_vl", "text_config": { "hidden_size": 6144, "intermediate_size": 3072, "num_hidden_layers": 60, "num_attention_heads": 64, "num_key_value_heads": 4, "head_dim": 128, "vocab_size": 200064, "max_position_embeddings": 1048576, "rms_norm_eps": 1e-06, "use_gemma_norm": true, "attention_output_gate": false, "rope_theta": 10000000, "rotary_dim": 64, "partial_rotary_factor": 0.5, "hidden_act": "swigluoai", "use_qk_norm": true, "tie_word_embeddings": false, "dense_intermediate_size": 12288, "shared_intermediate_size": 3072, "num_local_experts": 128, "num_experts_per_tok": 4, "n_shared_experts": 1, "scoring_func": "sigmoid", "use_routing_bias": true, "moe_layer_freq": [ 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ], "qk_norm_type": "per_head", "swiglu_alpha": 1.702, "swiglu_limit": 7.0, "routed_scaling_factor": 2.0, "sparse_attention_config": { "use_sparse_attention": true, "sparse_index_dim": 128, "sparse_num_index_heads": 4, "sparse_topk_blocks": 16, "sparse_block_size": 128, "sparse_score_type": "max", "sparse_init_block": 0, "sparse_local_block": 1, "sparse_attention_freq": [ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ], "sparse_disable_index_value": [ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 ] }, "architectures": [ "MiniMaxM3SparseForCausalLM" ] }, "vision_config": { "hidden_size": 1280, "num_attention_heads": 16, "num_hidden_layers": 32, "intermediate_size": 5120, "patch_size": 14, "image_size": 672, "projection_dim": 6144, "position_embedding_type": "rope", "rope_mode": "3d", "rope_theta": 10000.0, "attention_dropout": 0.0, "hidden_act": "gelu", "initializer_factor": 1.0, "initializer_range": 0.02, "layer_norm_eps": 1e-05, "model_type": "clip_vision_model", "num_channels": 3, "vocab_size": 32000, "img_token_compression_config": { "image_token_compression_method": "patch_merge", "spatial_merge_size": 2, "temporal_patch_size": 2 }, "vision_segment_max_frames": 4 }, "img_token_compression_config": { "image_token_compression_method": "patch_merge", "spatial_merge_size": 2, "temporal_patch_size": 2 }, "image_grid_pinpoints": "[(336, 336), (336, 672), (336, 1008), (336, 1344), (336, 1680), (336, 2016), (672, 336), (672, 672), (672, 1008), (672, 1344), (672, 1680), (672, 2016), (1008, 336), (1008, 672), (1008, 1008), (1008, 1344), (1008, 1680), (1008, 2016), (1344, 336), (1344, 672), (1344, 1008), (1344, 1344), (1344, 1680), (1344, 2016), (1680, 336), (1680, 672), (1680, 1008), (1680, 1344), (1680, 1680), (1680, 2016), (2016, 336), (2016, 672), (2016, 1008), (2016, 1344), (2016, 1680), (2016, 2016)]", "image_seq_length": 576, "image_token_index": 200025, "video_token_index": 200026, "multimodal_projector_bias": true, "num_reward_heads": 0, "process_image_mode": "dynamic_res", "projector_hidden_act": "gelu", "vision_feature_layer": -1, "vision_feature_select_strategy": "full", "torch_dtype": "bfloat16", "transformers_version": "4.52.4", "projector_hidden_size": 6144, "quantization": { "group_size": 64, "bits": 8, "lm_head": { "group_size": 64, "bits": 6 }, "model.embed_tokens": { "group_size": 64, "bits": 6 }, "model.layers.0.mlp.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.0.mlp.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.0.mlp.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.0.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.0.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.0.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.0.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.0.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.0.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.1.mlp.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.1.mlp.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.1.mlp.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.1.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.1.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.1.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.1.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.1.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.1.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.10.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.10.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.10.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.10.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.10.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.10.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.10.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.10.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.10.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.10.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.10.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.10.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.11.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.11.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.11.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.11.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.11.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.11.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.11.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.11.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.11.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.11.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.11.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.11.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.12.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.12.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.12.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.12.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.12.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.12.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.12.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.12.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.12.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.12.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.12.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.12.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.13.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.13.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.13.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.13.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.13.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.13.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.13.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.13.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.13.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.13.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.13.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.13.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.14.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.14.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.14.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.14.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.14.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.14.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.14.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.14.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.14.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.14.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.14.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.14.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.15.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.15.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.15.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.15.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.15.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.15.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.15.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.15.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.15.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.15.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.15.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.15.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.16.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.16.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.16.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.16.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.16.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.16.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.16.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.16.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.16.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.16.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.16.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.16.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.17.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.17.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.17.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.17.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.17.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.17.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.17.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.17.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.17.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.17.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.17.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.17.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.18.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.18.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.18.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.18.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.18.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.18.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.18.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.18.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.18.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.18.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.18.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.18.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.19.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.19.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.19.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.19.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.19.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.19.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.19.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.19.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.19.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.19.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.19.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.19.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.2.mlp.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.2.mlp.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.2.mlp.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.2.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.2.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.2.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.2.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.2.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.2.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.20.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.20.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.20.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.20.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.20.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.20.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.20.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.20.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.20.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.20.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.20.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.20.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.21.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.21.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.21.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.21.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.21.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.21.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.21.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.21.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.21.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.21.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.21.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.21.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.22.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.22.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.22.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.22.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.22.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.22.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.22.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.22.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.22.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.22.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.22.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.22.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.23.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.23.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.23.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.23.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.23.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.23.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.23.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.23.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.23.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.23.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.23.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.23.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.24.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.24.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.24.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.24.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.24.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.24.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.24.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.24.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.24.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.24.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.24.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.24.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.25.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.25.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.25.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.25.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.25.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.25.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.25.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.25.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.25.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.25.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.25.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.25.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.26.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.26.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.26.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.26.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.26.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.26.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.26.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.26.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.26.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.26.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.26.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.26.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.27.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.27.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.27.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.27.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.27.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.27.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.27.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.27.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.27.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.27.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.27.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.27.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.28.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.28.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.28.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.28.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.28.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.28.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.28.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.28.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.28.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.28.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.28.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.28.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.29.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.29.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.29.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.29.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.29.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.29.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.29.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.29.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.29.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.29.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.29.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.29.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.3.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.3.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.3.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.3.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.3.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.3.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.3.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.3.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.3.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.3.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.3.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.3.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.30.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.30.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.30.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.30.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.30.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.30.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.30.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.30.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.30.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.30.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.30.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.30.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.31.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.31.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.31.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.31.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.31.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.31.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.31.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.31.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.31.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.31.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.31.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.31.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.32.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.32.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.32.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.32.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.32.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.32.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.32.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.32.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.32.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.32.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.32.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.32.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.33.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.33.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.33.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.33.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.33.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.33.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.33.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.33.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.33.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.33.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.33.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.33.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.34.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.34.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.34.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.34.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.34.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.34.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.34.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.34.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.34.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.34.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.34.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.34.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.35.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.35.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.35.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.35.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.35.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.35.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.35.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.35.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.35.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.35.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.35.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.35.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.36.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.36.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.36.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.36.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.36.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.36.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.36.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.36.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.36.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.36.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.36.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.36.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.37.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.37.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.37.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.37.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.37.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.37.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.37.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.37.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.37.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.37.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.37.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.37.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.38.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.38.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.38.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.38.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.38.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.38.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.38.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.38.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.38.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.38.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.38.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.38.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.39.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.39.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.39.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.39.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.39.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.39.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.39.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.39.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.39.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.39.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.39.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.39.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.4.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.4.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.4.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.4.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.4.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.4.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.4.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.4.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.4.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.4.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.4.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.4.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.40.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.40.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.40.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.40.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.40.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.40.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.40.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.40.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.40.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.40.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.40.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.40.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.41.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.41.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.41.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.41.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.41.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.41.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.41.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.41.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.41.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.41.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.41.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.41.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.42.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.42.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.42.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.42.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.42.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.42.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.42.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.42.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.42.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.42.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.42.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.42.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.43.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.43.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.43.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.43.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.43.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.43.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.43.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.43.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.43.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.43.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.43.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.43.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.44.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.44.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.44.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.44.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.44.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.44.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.44.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.44.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.44.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.44.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.44.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.44.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.45.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.45.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.45.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.45.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.45.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.45.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.45.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.45.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.45.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.45.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.45.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.45.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.46.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.46.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.46.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.46.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.46.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.46.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.46.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.46.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.46.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.46.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.46.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.46.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.47.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.47.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.47.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.47.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.47.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.47.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.47.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.47.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.47.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.47.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.47.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.47.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.48.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.48.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.48.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.48.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.48.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.48.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.48.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.48.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.48.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.48.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.48.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.48.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.49.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.49.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.49.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.49.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.49.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.49.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.49.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.49.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.49.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.49.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.49.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.49.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.5.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.5.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.5.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.5.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.5.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.5.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.5.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.5.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.5.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.5.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.5.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.5.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.50.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.50.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.50.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.50.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.50.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.50.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.50.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.50.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.50.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.50.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.50.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.50.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.51.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.51.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.51.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.51.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.51.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.51.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.51.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.51.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.51.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.51.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.51.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.51.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.52.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.52.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.52.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.52.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.52.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.52.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.52.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.52.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.52.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.52.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.52.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.52.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.53.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.53.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.53.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.53.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.53.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.53.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.53.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.53.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.53.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.53.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.53.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.53.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.54.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.54.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.54.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.54.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.54.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.54.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.54.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.54.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.54.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.54.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.54.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.54.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.55.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.55.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.55.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.55.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.55.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.55.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.55.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.55.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.55.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.55.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.55.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.55.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.56.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.56.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.56.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.56.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.56.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.56.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.56.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.56.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.56.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.56.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.56.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.56.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.57.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.57.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.57.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.57.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.57.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.57.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.57.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.57.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.57.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.57.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.57.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.57.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.58.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.58.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.58.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.58.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.58.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.58.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.58.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.58.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.58.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.58.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.58.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.58.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.59.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.59.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.59.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.59.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.5 }, "model.layers.59.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.59.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.59.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.59.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.59.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.59.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.59.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.59.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.6.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.6.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.6.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.6.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 2.0 }, "model.layers.6.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.6.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.6.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.6.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.6.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.6.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.6.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.6.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.7.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.7.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.7.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.7.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.7.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.7.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.7.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.7.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.7.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.7.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.7.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.7.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.8.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.8.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.8.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.8.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.8.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.8.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.8.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.8.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.8.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.8.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.8.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.8.self_attn.v_proj": { "group_size": 64, "bits": 4 }, "model.layers.9.mlp.shared_experts.down_proj": { "group_size": 64, "bits": 4 }, "model.layers.9.mlp.shared_experts.gate_proj": { "group_size": 64, "bits": 4 }, "model.layers.9.mlp.shared_experts.up_proj": { "group_size": 64, "bits": 4 }, "model.layers.9.mlp.switch_mlp.down_proj": { "mode": "jangt", "kbits": 1.5 }, "model.layers.9.mlp.switch_mlp.gate_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.9.mlp.switch_mlp.up_proj": { "mode": "jangt", "kbits": 1.0 }, "model.layers.9.self_attn.indexer.index_k_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.9.self_attn.indexer.index_q_proj": { "mode": "mxfp8", "group_size": 32, "bits": 8 }, "model.layers.9.self_attn.k_proj": { "group_size": 64, "bits": 4 }, "model.layers.9.self_attn.o_proj": { "group_size": 64, "bits": 4 }, "model.layers.9.self_attn.q_proj": { "group_size": 64, "bits": 4 }, "model.layers.9.self_attn.v_proj": { "group_size": 64, "bits": 4 } }, "jangt": { "version": 1, "code": "v2_halfbits", "state_bits": 12, "rotation": "hadamard128-signs", "outlier_columns": { "model.layers.10.mlp.switch_mlp.down_proj": 8, "model.layers.10.mlp.switch_mlp.gate_proj": 1, "model.layers.10.mlp.switch_mlp.up_proj": 1, "model.layers.11.mlp.switch_mlp.down_proj": 8, "model.layers.11.mlp.switch_mlp.gate_proj": 1, "model.layers.11.mlp.switch_mlp.up_proj": 1, "model.layers.12.mlp.switch_mlp.down_proj": 8, "model.layers.12.mlp.switch_mlp.gate_proj": 1, "model.layers.12.mlp.switch_mlp.up_proj": 1, "model.layers.13.mlp.switch_mlp.down_proj": 7, "model.layers.13.mlp.switch_mlp.gate_proj": 1, "model.layers.13.mlp.switch_mlp.up_proj": 1, "model.layers.14.mlp.switch_mlp.down_proj": 8, "model.layers.14.mlp.switch_mlp.gate_proj": 0, "model.layers.14.mlp.switch_mlp.up_proj": 0, "model.layers.15.mlp.switch_mlp.down_proj": 8, "model.layers.15.mlp.switch_mlp.gate_proj": 0, "model.layers.15.mlp.switch_mlp.up_proj": 0, "model.layers.16.mlp.switch_mlp.down_proj": 6, "model.layers.16.mlp.switch_mlp.gate_proj": 0, "model.layers.16.mlp.switch_mlp.up_proj": 0, "model.layers.17.mlp.switch_mlp.down_proj": 8, "model.layers.17.mlp.switch_mlp.gate_proj": 0, "model.layers.17.mlp.switch_mlp.up_proj": 0, "model.layers.18.mlp.switch_mlp.down_proj": 8, "model.layers.18.mlp.switch_mlp.gate_proj": 0, "model.layers.18.mlp.switch_mlp.up_proj": 0, "model.layers.19.mlp.switch_mlp.down_proj": 8, "model.layers.19.mlp.switch_mlp.gate_proj": 0, "model.layers.19.mlp.switch_mlp.up_proj": 0, "model.layers.20.mlp.switch_mlp.down_proj": 8, "model.layers.20.mlp.switch_mlp.gate_proj": 0, "model.layers.20.mlp.switch_mlp.up_proj": 0, "model.layers.21.mlp.switch_mlp.down_proj": 8, "model.layers.21.mlp.switch_mlp.gate_proj": 0, "model.layers.21.mlp.switch_mlp.up_proj": 0, "model.layers.22.mlp.switch_mlp.down_proj": 8, "model.layers.22.mlp.switch_mlp.gate_proj": 0, "model.layers.22.mlp.switch_mlp.up_proj": 0, "model.layers.23.mlp.switch_mlp.down_proj": 8, "model.layers.23.mlp.switch_mlp.gate_proj": 0, "model.layers.23.mlp.switch_mlp.up_proj": 0, "model.layers.24.mlp.switch_mlp.down_proj": 8, "model.layers.24.mlp.switch_mlp.gate_proj": 0, "model.layers.24.mlp.switch_mlp.up_proj": 0, "model.layers.25.mlp.switch_mlp.down_proj": 8, "model.layers.25.mlp.switch_mlp.gate_proj": 0, "model.layers.25.mlp.switch_mlp.up_proj": 0, "model.layers.26.mlp.switch_mlp.down_proj": 8, "model.layers.26.mlp.switch_mlp.gate_proj": 0, "model.layers.26.mlp.switch_mlp.up_proj": 0, "model.layers.27.mlp.switch_mlp.down_proj": 8, "model.layers.27.mlp.switch_mlp.gate_proj": 0, "model.layers.27.mlp.switch_mlp.up_proj": 0, "model.layers.28.mlp.switch_mlp.down_proj": 8, "model.layers.28.mlp.switch_mlp.gate_proj": 0, "model.layers.28.mlp.switch_mlp.up_proj": 0, "model.layers.29.mlp.switch_mlp.down_proj": 8, "model.layers.29.mlp.switch_mlp.gate_proj": 0, "model.layers.29.mlp.switch_mlp.up_proj": 0, "model.layers.3.mlp.switch_mlp.down_proj": 8, "model.layers.3.mlp.switch_mlp.gate_proj": 2, "model.layers.3.mlp.switch_mlp.up_proj": 2, "model.layers.30.mlp.switch_mlp.down_proj": 8, "model.layers.30.mlp.switch_mlp.gate_proj": 0, "model.layers.30.mlp.switch_mlp.up_proj": 0, "model.layers.31.mlp.switch_mlp.down_proj": 8, "model.layers.31.mlp.switch_mlp.gate_proj": 0, "model.layers.31.mlp.switch_mlp.up_proj": 0, "model.layers.32.mlp.switch_mlp.down_proj": 8, "model.layers.32.mlp.switch_mlp.gate_proj": 0, "model.layers.32.mlp.switch_mlp.up_proj": 0, "model.layers.33.mlp.switch_mlp.down_proj": 8, "model.layers.33.mlp.switch_mlp.gate_proj": 0, "model.layers.33.mlp.switch_mlp.up_proj": 0, "model.layers.34.mlp.switch_mlp.down_proj": 6, "model.layers.34.mlp.switch_mlp.gate_proj": 0, "model.layers.34.mlp.switch_mlp.up_proj": 0, "model.layers.35.mlp.switch_mlp.down_proj": 8, "model.layers.35.mlp.switch_mlp.gate_proj": 0, "model.layers.35.mlp.switch_mlp.up_proj": 0, "model.layers.36.mlp.switch_mlp.down_proj": 8, "model.layers.36.mlp.switch_mlp.gate_proj": 0, "model.layers.36.mlp.switch_mlp.up_proj": 0, "model.layers.37.mlp.switch_mlp.down_proj": 8, "model.layers.37.mlp.switch_mlp.gate_proj": 0, "model.layers.37.mlp.switch_mlp.up_proj": 0, "model.layers.38.mlp.switch_mlp.down_proj": 8, "model.layers.38.mlp.switch_mlp.gate_proj": 0, "model.layers.38.mlp.switch_mlp.up_proj": 0, "model.layers.39.mlp.switch_mlp.down_proj": 8, "model.layers.39.mlp.switch_mlp.gate_proj": 0, "model.layers.39.mlp.switch_mlp.up_proj": 0, "model.layers.4.mlp.switch_mlp.down_proj": 8, "model.layers.4.mlp.switch_mlp.gate_proj": 1, "model.layers.4.mlp.switch_mlp.up_proj": 1, "model.layers.40.mlp.switch_mlp.down_proj": 8, "model.layers.40.mlp.switch_mlp.gate_proj": 0, "model.layers.40.mlp.switch_mlp.up_proj": 0, "model.layers.41.mlp.switch_mlp.down_proj": 3, "model.layers.41.mlp.switch_mlp.gate_proj": 0, "model.layers.41.mlp.switch_mlp.up_proj": 0, "model.layers.42.mlp.switch_mlp.down_proj": 8, "model.layers.42.mlp.switch_mlp.gate_proj": 0, "model.layers.42.mlp.switch_mlp.up_proj": 0, "model.layers.43.mlp.switch_mlp.down_proj": 6, "model.layers.43.mlp.switch_mlp.gate_proj": 0, "model.layers.43.mlp.switch_mlp.up_proj": 0, "model.layers.44.mlp.switch_mlp.down_proj": 8, "model.layers.44.mlp.switch_mlp.gate_proj": 0, "model.layers.44.mlp.switch_mlp.up_proj": 0, "model.layers.45.mlp.switch_mlp.down_proj": 8, "model.layers.45.mlp.switch_mlp.gate_proj": 0, "model.layers.45.mlp.switch_mlp.up_proj": 0, "model.layers.46.mlp.switch_mlp.down_proj": 8, "model.layers.46.mlp.switch_mlp.gate_proj": 0, "model.layers.46.mlp.switch_mlp.up_proj": 0, "model.layers.47.mlp.switch_mlp.down_proj": 3, "model.layers.47.mlp.switch_mlp.gate_proj": 0, "model.layers.47.mlp.switch_mlp.up_proj": 0, "model.layers.48.mlp.switch_mlp.down_proj": 4, "model.layers.48.mlp.switch_mlp.gate_proj": 0, "model.layers.48.mlp.switch_mlp.up_proj": 0, "model.layers.49.mlp.switch_mlp.down_proj": 8, "model.layers.49.mlp.switch_mlp.gate_proj": 0, "model.layers.49.mlp.switch_mlp.up_proj": 0, "model.layers.5.mlp.switch_mlp.down_proj": 8, "model.layers.5.mlp.switch_mlp.gate_proj": 1, "model.layers.5.mlp.switch_mlp.up_proj": 1, "model.layers.50.mlp.switch_mlp.down_proj": 4, "model.layers.50.mlp.switch_mlp.gate_proj": 0, "model.layers.50.mlp.switch_mlp.up_proj": 0, "model.layers.51.mlp.switch_mlp.down_proj": 8, "model.layers.51.mlp.switch_mlp.gate_proj": 0, "model.layers.51.mlp.switch_mlp.up_proj": 0, "model.layers.52.mlp.switch_mlp.down_proj": 8, "model.layers.52.mlp.switch_mlp.gate_proj": 0, "model.layers.52.mlp.switch_mlp.up_proj": 0, "model.layers.53.mlp.switch_mlp.down_proj": 8, "model.layers.53.mlp.switch_mlp.gate_proj": 0, "model.layers.53.mlp.switch_mlp.up_proj": 0, "model.layers.54.mlp.switch_mlp.down_proj": 8, "model.layers.54.mlp.switch_mlp.gate_proj": 0, "model.layers.54.mlp.switch_mlp.up_proj": 0, "model.layers.55.mlp.switch_mlp.down_proj": 8, "model.layers.55.mlp.switch_mlp.gate_proj": 0, "model.layers.55.mlp.switch_mlp.up_proj": 0, "model.layers.56.mlp.switch_mlp.down_proj": 8, "model.layers.56.mlp.switch_mlp.gate_proj": 0, "model.layers.56.mlp.switch_mlp.up_proj": 0, "model.layers.57.mlp.switch_mlp.down_proj": 8, "model.layers.57.mlp.switch_mlp.gate_proj": 0, "model.layers.57.mlp.switch_mlp.up_proj": 0, "model.layers.58.mlp.switch_mlp.down_proj": 8, "model.layers.58.mlp.switch_mlp.gate_proj": 0, "model.layers.58.mlp.switch_mlp.up_proj": 0, "model.layers.59.mlp.switch_mlp.down_proj": 8, "model.layers.59.mlp.switch_mlp.gate_proj": 0, "model.layers.59.mlp.switch_mlp.up_proj": 0, "model.layers.6.mlp.switch_mlp.down_proj": 8, "model.layers.6.mlp.switch_mlp.gate_proj": 1, "model.layers.6.mlp.switch_mlp.up_proj": 1, "model.layers.7.mlp.switch_mlp.down_proj": 8, "model.layers.7.mlp.switch_mlp.gate_proj": 1, "model.layers.7.mlp.switch_mlp.up_proj": 1, "model.layers.8.mlp.switch_mlp.down_proj": 8, "model.layers.8.mlp.switch_mlp.gate_proj": 1, "model.layers.8.mlp.switch_mlp.up_proj": 1, "model.layers.9.mlp.switch_mlp.down_proj": 7, "model.layers.9.mlp.switch_mlp.gate_proj": 1, "model.layers.9.mlp.switch_mlp.up_proj": 1 } }, "jang_config": { "format": "jangtq2", "format_version": 2, "format_name": "JANGHT", "model_name": "MiniMax-M3.1-Flash-JANGHT1.9", "bpw": 1.878, "source": "MiniMaxAI/MiniMax-M3.1-Flash", "source_params": 427051937408, "expert_widths": { "3": { "gu": 1.0, "dn": 1.0 }, "4": { "gu": 1.0, "dn": 1.0 }, "5": { "gu": 1.0, "dn": 1.5 }, "6": { "gu": 1.0, "dn": 2.0 }, "7": { "gu": 1.0, "dn": 1.5 }, "8": { "gu": 1.0, "dn": 1.5 }, "9": { "gu": 1.0, "dn": 1.5 }, "10": { "gu": 1.0, "dn": 1.5 }, "11": { "gu": 1.0, "dn": 1.5 }, "12": { "gu": 1.0, "dn": 1.5 }, "13": { "gu": 1.0, "dn": 1.5 }, "14": { "gu": 1.0, "dn": 1.5 }, "15": { "gu": 1.0, "dn": 1.5 }, "16": { "gu": 1.0, "dn": 1.5 }, "17": { "gu": 1.0, "dn": 1.5 }, "18": { "gu": 1.0, "dn": 2.0 }, "19": { "gu": 1.5, "dn": 2.0 }, "20": { "gu": 1.5, "dn": 2.0 }, "21": { "gu": 1.5, "dn": 2.0 }, "22": { "gu": 1.5, "dn": 2.0 }, "23": { "gu": 1.5, "dn": 2.0 }, "24": { "gu": 1.5, "dn": 2.5 }, "25": { "gu": 1.5, "dn": 2.0 }, "26": { "gu": 1.5, "dn": 2.0 }, "27": { "gu": 1.5, "dn": 2.5 }, "28": { "gu": 1.5, "dn": 2.5 }, "29": { "gu": 1.5, "dn": 2.5 }, "30": { "gu": 1.5, "dn": 2.0 }, "31": { "gu": 1.5, "dn": 2.5 }, "32": { "gu": 1.5, "dn": 2.5 }, "33": { "gu": 1.5, "dn": 2.5 }, "34": { "gu": 1.5, "dn": 2.5 }, "35": { "gu": 1.5, "dn": 2.5 }, "36": { "gu": 1.5, "dn": 2.5 }, "37": { "gu": 1.5, "dn": 2.5 }, "38": { "gu": 1.5, "dn": 2.5 }, "39": { "gu": 2.0, "dn": 2.5 }, "40": { "gu": 2.0, "dn": 2.5 }, "41": { "gu": 1.5, "dn": 2.5 }, "42": { "gu": 1.5, "dn": 2.5 }, "43": { "gu": 2.0, "dn": 2.5 }, "44": { "gu": 2.0, "dn": 2.5 }, "45": { "gu": 1.5, "dn": 2.5 }, "46": { "gu": 2.0, "dn": 2.5 }, "47": { "gu": 1.5, "dn": 2.5 }, "48": { "gu": 2.0, "dn": 2.5 }, "49": { "gu": 2.0, "dn": 2.5 }, "50": { "gu": 2.0, "dn": 2.5 }, "51": { "gu": 2.0, "dn": 2.5 }, "52": { "gu": 2.0, "dn": 2.5 }, "53": { "gu": 2.0, "dn": 2.5 }, "54": { "gu": 2.0, "dn": 2.5 }, "55": { "gu": 2.0, "dn": 2.5 }, "56": { "gu": 2.0, "dn": 2.5 }, "57": { "gu": 2.0, "dn": 2.5 }, "58": { "gu": 1.5, "dn": 2.5 }, "59": { "gu": 1.5, "dn": 2.5 } }, "trunk": { "qo": "a4", "kv": "a4", "idx": "mxfp8", "shared": "a4", "dense": "a4" }, "embed_bits": 6, "head_bits": 6, "drafter": "dspark/ as released (MXFP8 layers, BF16 fc / Markov head)", "chat": { "sampling_defaults": { "temperature": 1.0, "top_p": 0.95 } } } }