{ "architectures": [ "MuseGlimmerForConditionalGeneration" ], "dtype": "bfloat16", "image_token_id": 200092, "model_type": "muse_glimmer", "out_hidden_size": 6144, "projector_hidden_act": "gelu", "projector_hidden_size": 4096, "text_config": { "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 200000, "eos_token_id": 200001, "final_logit_softcapping": 20.0, "head_dim": 128, "hidden_activation": "silu", "hidden_size": 6656, "initializer_range": 0.02, "intermediate_size": 19968, "layer_rope_theta": [ 500000.0, 500000.0, 500000.0, 0, 500000.0, 500000.0, 500000.0, 0, 500000.0, 500000.0, 500000.0, 0, 500000.0, 500000.0, 500000.0, 0, 500000.0, 500000.0, 500000.0, 0, 500000.0, 500000.0, 500000.0, 0, 500000.0, 500000.0, 500000.0, 0, 500000.0, 500000.0, 500000.0, 0, 500000.0, 500000.0, 500000.0, 0, 500000.0, 500000.0, 500000.0, 0, 500000.0, 500000.0, 500000.0, 0, 500000.0, 500000.0, 500000.0, 0, 500000.0, 500000.0, 500000.0, 0 ], "layer_types": [ "sliding_attention", "sliding_attention", "sliding_attention", "full_attention", "sliding_attention", "sliding_attention", "sliding_attention", "full_attention", "sliding_attention", "sliding_attention", "sliding_attention", "full_attention", "sliding_attention", "sliding_attention", "sliding_attention", "full_attention", "sliding_attention", "sliding_attention", "sliding_attention", "full_attention", "sliding_attention", "sliding_attention", "sliding_attention", "full_attention", "sliding_attention", "sliding_attention", "sliding_attention", "full_attention", "sliding_attention", "sliding_attention", "sliding_attention", "full_attention", "sliding_attention", "sliding_attention", "sliding_attention", "full_attention", "sliding_attention", "sliding_attention", "sliding_attention", "full_attention", "sliding_attention", "sliding_attention", "sliding_attention", "full_attention", "sliding_attention", "sliding_attention", "sliding_attention", "full_attention", "sliding_attention", "sliding_attention", "sliding_attention", "full_attention" ], "max_position_embeddings": 131072, "model_type": "muse_glimmer_text", "num_attention_heads": 32, "num_hidden_layers": 52, "num_key_value_heads": 2, "output_multiplier": 0.19611613513818404, "pad_token_id": null, "post_norm_eps": 1e-08, "qk_scale_factor": 3.87, "rms_norm_eps": 1e-05, "rope_parameters": { "rope_theta": 500000.0, "rope_type": "default" }, "sliding_window": 2048, "tie_word_embeddings": false, "use_cache": true, "vocab_size": 202048 }, "transformers_version": "5.15.0.dev0", "video_token_id": 200091, "vision_config": { "hidden_act": "gelu", "hidden_size": 1536, "intermediate_size": 8960, "layer_norm_eps": 1e-05, "layer_types": [ "window_attention", "window_attention", "window_attention", "full_attention", "window_attention", "window_attention", "window_attention", "full_attention", "window_attention", "window_attention", "window_attention", "full_attention", "window_attention", "window_attention", "window_attention", "full_attention", "window_attention", "window_attention", "window_attention", "full_attention", "window_attention", "window_attention", "window_attention", "full_attention", "window_attention", "window_attention", "window_attention", "full_attention", "window_attention", "window_attention", "window_attention", "full_attention", "window_attention", "window_attention", "window_attention", "full_attention", "window_attention", "window_attention", "window_attention", "full_attention", "window_attention", "window_attention", "window_attention", "full_attention", "window_attention", "window_attention", "window_attention", "full_attention", "window_attention", "full_attention" ], "max_position_embeddings": 1024, "merge_size": 2, "model_type": "muse_glimmer_vision", "num_attention_heads": 16, "num_hidden_layers": 50, "patch_size": 14, "patch_temporal": 2, "pos_emb_height": 32, "pos_emb_width": 32, "rope_parameters": { "rope_theta": 10000.0, "rope_type": "default" } }, "quantization_config": { "quant_algo": "MIXED_PRECISION", "quantized_layers": { "model.embed_tokens": { "quant_algo": "NVFP4", "group_size": 16 }, "lm_head": { "quant_algo": "MXFP8" }, "model.layers.0.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.0.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.0.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.0.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.0.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.0.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.0.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.1.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.1.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.1.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.1.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.1.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.1.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.1.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.2.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.2.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.2.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.2.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.2.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.2.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.2.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.3.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.3.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.3.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.3.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.3.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.3.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.3.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.4.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.4.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.4.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.4.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.4.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.4.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.4.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.5.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.5.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.5.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.5.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.5.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.5.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.5.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.6.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.6.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.6.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.6.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.6.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.6.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.6.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.7.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.7.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.7.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.7.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.7.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.7.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.7.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.8.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.8.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.8.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.8.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.8.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.8.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.8.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.9.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.9.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.9.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.9.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.9.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.9.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.9.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.10.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.10.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.10.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.10.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.10.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.10.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.10.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.11.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.11.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.11.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.11.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.11.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.11.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.11.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.12.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.12.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.12.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.12.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.12.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.12.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.12.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.13.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.13.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.13.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.13.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.13.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.13.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.13.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.14.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.14.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.14.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.14.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.14.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.14.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.14.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.15.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.15.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.15.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.15.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.15.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.15.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.15.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.16.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.16.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.16.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.16.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.16.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.16.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.16.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.17.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.17.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.17.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.17.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.17.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.17.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.17.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.18.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.18.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.18.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.18.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.18.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.18.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.18.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.19.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.19.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.19.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.19.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.19.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.19.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.19.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.20.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.20.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.20.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.20.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.20.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.20.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.20.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.21.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.21.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.21.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.21.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.21.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.21.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.21.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.22.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.22.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.22.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.22.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.22.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.22.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.22.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.23.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.23.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.23.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.23.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.23.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.23.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.23.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.24.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.24.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.24.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.24.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.24.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.24.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.24.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.25.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.25.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.25.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.25.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.25.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.25.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.25.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.26.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.26.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.26.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.26.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.26.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.26.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.26.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.27.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.27.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.27.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.27.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.27.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.27.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.27.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.28.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.28.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.28.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.28.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.28.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.28.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.28.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.29.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.29.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.29.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.29.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.29.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.29.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.29.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.30.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.30.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.30.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.30.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.30.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.30.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.30.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.31.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.31.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.31.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.31.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.31.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.31.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.31.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.32.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.32.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.32.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.32.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.32.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.32.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.32.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.33.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.33.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.33.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.33.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.33.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.33.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.33.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.34.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.34.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.34.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.34.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.34.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.34.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.34.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.35.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.35.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.35.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.35.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.35.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.35.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.35.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.36.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.36.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.36.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.36.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.36.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.36.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.36.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.37.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.37.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.37.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.37.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.37.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.37.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.37.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.38.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.38.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.38.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.38.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.38.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.38.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.38.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.39.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.39.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.39.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.39.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.39.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.39.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.39.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.40.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.40.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.40.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.40.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.40.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.40.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.40.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.41.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.41.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.41.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.41.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.41.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.41.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.41.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.42.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.42.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.42.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.42.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.42.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.42.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.42.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.43.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.43.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.43.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.43.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.43.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.43.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.43.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.44.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.44.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.44.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.44.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.44.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.44.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.44.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.45.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.45.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.45.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.45.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.45.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.45.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.45.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.46.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.46.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.46.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.46.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.46.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.46.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.46.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.47.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.47.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.47.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.47.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.47.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.47.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.47.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.48.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.48.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.48.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.48.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.48.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.48.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.48.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.49.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.49.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.49.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.49.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.49.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.49.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.49.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.50.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.50.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.50.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.50.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.50.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.50.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.50.mlp.down_proj": { "quant_algo": "MXFP8" }, "model.layers.51.self_attn.q_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.51.self_attn.k_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.51.self_attn.output_gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.51.mlp.gate_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.51.mlp.up_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.51.self_attn.o_proj": { "quant_algo": "NVFP4", "group_size": 16 }, "model.layers.51.mlp.down_proj": { "quant_algo": "MXFP8" } }, "ignore": [], "exclude_modules": [] } }