| { |
| "producer": { |
| "name": "modelopt", |
| "cast": "mxfp4_to_nvfp4_offline" |
| }, |
| "quantization": { |
| "quant_algo": "NVFP4", |
| "kv_cache_quant_algo": null, |
| "group_size": 16, |
| "exclude_modules": [ |
| "*self_attn*", |
| "*shared_experts*", |
| "*mlp.gate_proj*", |
| "*mlp.up_proj*", |
| "*mlp.gate_up_proj*", |
| "*mlp.down_proj*", |
| "*lm_head*", |
| "*vision_tower*", |
| "*mm_projector*", |
| "*language_model.lm_head*", |
| "*language_model.model.embed_tokens*", |
| "*input_layernorm*", |
| "*mlp_res_norm*", |
| "*mlp_res_proj*", |
| "*post_attention_layernorm*", |
| "*self_attention_res_norm*", |
| "*self_attention_res_proj*", |
| "*self_attn.b_proj*", |
| "*self_attn.f_a_proj*", |
| "*self_attn.f_b_proj*", |
| "*self_attn.g_proj*", |
| "*self_attn.k_conv1d*", |
| "*self_attn.k_proj*", |
| "*self_attn.o_norm*", |
| "*self_attn.o_proj*", |
| "*self_attn.q_conv1d*", |
| "*self_attn.q_proj*", |
| "*self_attn.v_conv1d*", |
| "*self_attn.v_proj*", |
| "*block_sparse_moe.gate*", |
| "*block_sparse_moe.routed_expert_down_proj*", |
| "*block_sparse_moe.routed_expert_norm*", |
| "*block_sparse_moe.routed_expert_up_proj*", |
| "*block_sparse_moe.shared_experts.down_proj*", |
| "*block_sparse_moe.shared_experts.gate_proj*", |
| "*block_sparse_moe.shared_experts.up_proj*", |
| "*self_attn.kv_a_layernorm*", |
| "*self_attn.kv_a_proj_with_mqa*", |
| "*self_attn.kv_b_proj*", |
| "*self_attn.q_a_layernorm*", |
| "*self_attn.q_a_proj*", |
| "*self_attn.q_b_proj*", |
| "*language_model.model.norm*", |
| "*language_model.model.output_attn_res_norm*", |
| "*language_model.model.output_attn_res_proj*", |
| "*mm_projector.post_norm*", |
| "*mlp.fc0*", |
| "*mlp.fc1*", |
| "*norm0*", |
| "*norm1*", |
| "*wo*", |
| "*wqkv*", |
| "*vision_tower.encoder.final_layernorm*", |
| "*vision_tower.patch_embed.pos_emb*", |
| "*vision_tower.patch_embed.proj*" |
| ], |
| "quantized_modules_count": 247296, |
| "activation_scheme": "none" |
| } |
| } |
