| { |
| "architectures": [ |
| "Qwen3_5MoeForCausalLM" |
| ], |
| "attention_bias": false, |
| "attention_dropout": 0.0, |
| "attn_output_gate": true, |
| "base_model_ep_plan": { |
| "layers.*.mlp.experts": "torch_a2a_experts", |
| "layers.*.mlp.experts.down_proj": "grouped_gemm", |
| "layers.*.mlp.experts.gate_up_proj": "grouped_gemm", |
| "layers.*.mlp.gate": "torch_a2a_router" |
| }, |
| "bos_token_id": 248044, |
| "distributed_config": null, |
| "dtype": "bfloat16", |
| "eos_token_id": 248044, |
| "full_attention_interval": 4, |
| "head_dim": 256, |
| "hidden_act": "silu", |
| "hidden_size": 8192, |
| "initializer_range": 0.02, |
| "layer_types": [ |
| "linear_attention", |
| "linear_attention", |
| "linear_attention", |
| "full_attention", |
| "linear_attention", |
| "linear_attention", |
| "linear_attention", |
| "full_attention", |
| "linear_attention", |
| "linear_attention", |
| "linear_attention", |
| "full_attention", |
| "linear_attention", |
| "linear_attention", |
| "linear_attention", |
| "full_attention", |
| "linear_attention", |
| "linear_attention", |
| "linear_attention", |
| "full_attention", |
| "linear_attention", |
| "linear_attention", |
| "linear_attention", |
| "full_attention", |
| "linear_attention", |
| "linear_attention", |
| "linear_attention", |
| "full_attention", |
| "linear_attention", |
| "linear_attention", |
| "linear_attention", |
| "full_attention", |
| "linear_attention", |
| "linear_attention", |
| "linear_attention", |
| "full_attention", |
| "linear_attention", |
| "linear_attention", |
| "linear_attention", |
| "full_attention", |
| "linear_attention", |
| "linear_attention", |
| "linear_attention", |
| "full_attention", |
| "linear_attention", |
| "linear_attention", |
| "linear_attention", |
| "full_attention", |
| "linear_attention", |
| "linear_attention", |
| "linear_attention", |
| "full_attention", |
| "linear_attention", |
| "linear_attention", |
| "linear_attention", |
| "full_attention", |
| "linear_attention", |
| "linear_attention", |
| "linear_attention", |
| "full_attention", |
| "linear_attention", |
| "linear_attention", |
| "linear_attention", |
| "full_attention", |
| "linear_attention", |
| "linear_attention", |
| "linear_attention", |
| "full_attention", |
| "linear_attention", |
| "linear_attention", |
| "linear_attention", |
| "full_attention", |
| "linear_attention", |
| "linear_attention", |
| "linear_attention", |
| "full_attention", |
| "linear_attention", |
| "linear_attention", |
| "linear_attention", |
| "full_attention", |
| "linear_attention", |
| "linear_attention", |
| "linear_attention", |
| "full_attention", |
| "linear_attention", |
| "linear_attention", |
| "linear_attention", |
| "full_attention", |
| "linear_attention", |
| "linear_attention", |
| "linear_attention", |
| "full_attention" |
| ], |
| "linear_conv_kernel_dim": 4, |
| "linear_key_head_dim": 128, |
| "linear_num_key_heads": 16, |
| "linear_num_value_heads": 128, |
| "linear_value_head_dim": 128, |
| "mamba_ssm_dtype": "float32", |
| "max_position_embeddings": 262144, |
| "model_type": "qwen3_5_moe_text", |
| "moe_intermediate_size": 2048, |
| "mtp_num_hidden_layers": 1, |
| "mtp_use_dedicated_embeddings": false, |
| "num_attention_heads": 64, |
| "num_experts": 512, |
| "num_experts_per_tok": 10, |
| "num_hidden_layers": 92, |
| "num_key_value_heads": 4, |
| "output_gate_type": "swish", |
| "output_router_logits": false, |
| "pad_token_id": null, |
| "partial_rotary_factor": 0.25, |
| "rms_norm_eps": 1e-06, |
| "rope_parameters": { |
| "mrope_interleaved": true, |
| "mrope_section": [ |
| 11, |
| 11, |
| 10 |
| ], |
| "partial_rotary_factor": 0.25, |
| "rope_theta": 10000000, |
| "rope_type": "default" |
| }, |
| "router_aux_loss_coef": 0.001, |
| "shared_expert_intermediate_size": 2048, |
| "tie_word_embeddings": false, |
| "transformers_version": "5.13.0", |
| "use_cache": true, |
| "vocab_size": 248320, |
| "quantization_config": { |
| "config_groups": { |
| "group_0": { |
| "input_activations": { |
| "dynamic": false, |
| "num_bits": 8, |
| "type": "float" |
| }, |
| "weights": { |
| "dynamic": false, |
| "num_bits": 8, |
| "type": "float" |
| }, |
| "targets": [ |
| "model.layers.0.linear_attn.conv1d", |
| "model.layers.0.linear_attn.in_proj_a", |
| "model.layers.0.linear_attn.in_proj_b", |
| "model.layers.0.linear_attn.in_proj_qkv", |
| "model.layers.0.linear_attn.in_proj_z", |
| "model.layers.0.linear_attn.out_proj", |
| "model.layers.1.linear_attn.conv1d", |
| "model.layers.1.linear_attn.in_proj_a", |
| "model.layers.1.linear_attn.in_proj_b", |
| "model.layers.1.linear_attn.in_proj_qkv", |
| "model.layers.1.linear_attn.in_proj_z", |
| "model.layers.1.linear_attn.out_proj", |
| "model.layers.10.linear_attn.conv1d", |
| "model.layers.10.linear_attn.in_proj_a", |
| "model.layers.10.linear_attn.in_proj_b", |
| "model.layers.10.linear_attn.in_proj_qkv", |
| "model.layers.10.linear_attn.in_proj_z", |
| "model.layers.10.linear_attn.out_proj", |
| "model.layers.11.self_attn.k_proj", |
| "model.layers.11.self_attn.o_proj", |
| "model.layers.11.self_attn.q_proj", |
| "model.layers.11.self_attn.v_proj", |
| "model.layers.12.linear_attn.conv1d", |
| "model.layers.12.linear_attn.in_proj_a", |
| "model.layers.12.linear_attn.in_proj_b", |
| "model.layers.12.linear_attn.in_proj_qkv", |
| "model.layers.12.linear_attn.in_proj_z", |
| "model.layers.12.linear_attn.out_proj", |
| "model.layers.13.linear_attn.conv1d", |
| "model.layers.13.linear_attn.in_proj_a", |
| "model.layers.13.linear_attn.in_proj_b", |
| "model.layers.13.linear_attn.in_proj_qkv", |
| "model.layers.13.linear_attn.in_proj_z", |
| "model.layers.13.linear_attn.out_proj", |
| "model.layers.14.linear_attn.conv1d", |
| "model.layers.14.linear_attn.in_proj_a", |
| "model.layers.14.linear_attn.in_proj_b", |
| "model.layers.14.linear_attn.in_proj_qkv", |
| "model.layers.14.linear_attn.in_proj_z", |
| "model.layers.14.linear_attn.out_proj", |
| "model.layers.15.self_attn.k_proj", |
| "model.layers.15.self_attn.o_proj", |
| "model.layers.15.self_attn.q_proj", |
| "model.layers.15.self_attn.v_proj", |
| "model.layers.16.linear_attn.conv1d", |
| "model.layers.16.linear_attn.in_proj_a", |
| "model.layers.16.linear_attn.in_proj_b", |
| "model.layers.16.linear_attn.in_proj_qkv", |
| "model.layers.16.linear_attn.in_proj_z", |
| "model.layers.16.linear_attn.out_proj", |
| "model.layers.17.linear_attn.conv1d", |
| "model.layers.17.linear_attn.in_proj_a", |
| "model.layers.17.linear_attn.in_proj_b", |
| "model.layers.17.linear_attn.in_proj_qkv", |
| "model.layers.17.linear_attn.in_proj_z", |
| "model.layers.17.linear_attn.out_proj", |
| "model.layers.18.linear_attn.conv1d", |
| "model.layers.18.linear_attn.in_proj_a", |
| "model.layers.18.linear_attn.in_proj_b", |
| "model.layers.18.linear_attn.in_proj_qkv", |
| "model.layers.18.linear_attn.in_proj_z", |
| "model.layers.18.linear_attn.out_proj", |
| "model.layers.19.self_attn.k_proj", |
| "model.layers.19.self_attn.o_proj", |
| "model.layers.19.self_attn.q_proj", |
| "model.layers.19.self_attn.v_proj", |
| "model.layers.2.linear_attn.conv1d", |
| "model.layers.2.linear_attn.in_proj_a", |
| "model.layers.2.linear_attn.in_proj_b", |
| "model.layers.2.linear_attn.in_proj_qkv", |
| "model.layers.2.linear_attn.in_proj_z", |
| "model.layers.2.linear_attn.out_proj", |
| "model.layers.20.linear_attn.conv1d", |
| "model.layers.20.linear_attn.in_proj_a", |
| "model.layers.20.linear_attn.in_proj_b", |
| "model.layers.20.linear_attn.in_proj_qkv", |
| "model.layers.20.linear_attn.in_proj_z", |
| "model.layers.20.linear_attn.out_proj", |
| "model.layers.21.linear_attn.conv1d", |
| "model.layers.21.linear_attn.in_proj_a", |
| "model.layers.21.linear_attn.in_proj_b", |
| "model.layers.21.linear_attn.in_proj_qkv", |
| "model.layers.21.linear_attn.in_proj_z", |
| "model.layers.21.linear_attn.out_proj", |
| "model.layers.22.linear_attn.conv1d", |
| "model.layers.22.linear_attn.in_proj_a", |
| "model.layers.22.linear_attn.in_proj_b", |
| "model.layers.22.linear_attn.in_proj_qkv", |
| "model.layers.22.linear_attn.in_proj_z", |
| "model.layers.22.linear_attn.out_proj", |
| "model.layers.23.self_attn.k_proj", |
| "model.layers.23.self_attn.o_proj", |
| "model.layers.23.self_attn.q_proj", |
| "model.layers.23.self_attn.v_proj", |
| "model.layers.24.linear_attn.conv1d", |
| "model.layers.24.linear_attn.in_proj_a", |
| "model.layers.24.linear_attn.in_proj_b", |
| "model.layers.24.linear_attn.in_proj_qkv", |
| "model.layers.24.linear_attn.in_proj_z", |
| "model.layers.24.linear_attn.out_proj", |
| "model.layers.25.linear_attn.conv1d", |
| "model.layers.25.linear_attn.in_proj_a", |
| "model.layers.25.linear_attn.in_proj_b", |
| "model.layers.25.linear_attn.in_proj_qkv", |
| "model.layers.25.linear_attn.in_proj_z", |
| "model.layers.25.linear_attn.out_proj", |
| "model.layers.26.linear_attn.conv1d", |
| "model.layers.26.linear_attn.in_proj_a", |
| "model.layers.26.linear_attn.in_proj_b", |
| "model.layers.26.linear_attn.in_proj_qkv", |
| "model.layers.26.linear_attn.in_proj_z", |
| "model.layers.26.linear_attn.out_proj", |
| "model.layers.27.self_attn.k_proj", |
| "model.layers.27.self_attn.o_proj", |
| "model.layers.27.self_attn.q_proj", |
| "model.layers.27.self_attn.v_proj", |
| "model.layers.28.linear_attn.conv1d", |
| "model.layers.28.linear_attn.in_proj_a", |
| "model.layers.28.linear_attn.in_proj_b", |
| "model.layers.28.linear_attn.in_proj_qkv", |
| "model.layers.28.linear_attn.in_proj_z", |
| "model.layers.28.linear_attn.out_proj", |
| "model.layers.29.linear_attn.conv1d", |
| "model.layers.29.linear_attn.in_proj_a", |
| "model.layers.29.linear_attn.in_proj_b", |
| "model.layers.29.linear_attn.in_proj_qkv", |
| "model.layers.29.linear_attn.in_proj_z", |
| "model.layers.29.linear_attn.out_proj", |
| "model.layers.3.self_attn.k_proj", |
| "model.layers.3.self_attn.o_proj", |
| "model.layers.3.self_attn.q_proj", |
| "model.layers.3.self_attn.v_proj", |
| "model.layers.30.linear_attn.conv1d", |
| "model.layers.30.linear_attn.in_proj_a", |
| "model.layers.30.linear_attn.in_proj_b", |
| "model.layers.30.linear_attn.in_proj_qkv", |
| "model.layers.30.linear_attn.in_proj_z", |
| "model.layers.30.linear_attn.out_proj", |
| "model.layers.31.self_attn.k_proj", |
| "model.layers.31.self_attn.o_proj", |
| "model.layers.31.self_attn.q_proj", |
| "model.layers.31.self_attn.v_proj", |
| "model.layers.32.linear_attn.conv1d", |
| "model.layers.32.linear_attn.in_proj_a", |
| "model.layers.32.linear_attn.in_proj_b", |
| "model.layers.32.linear_attn.in_proj_qkv", |
| "model.layers.32.linear_attn.in_proj_z", |
| "model.layers.32.linear_attn.out_proj", |
| "model.layers.33.linear_attn.conv1d", |
| "model.layers.33.linear_attn.in_proj_a", |
| "model.layers.33.linear_attn.in_proj_b", |
| "model.layers.33.linear_attn.in_proj_qkv", |
| "model.layers.33.linear_attn.in_proj_z", |
| "model.layers.33.linear_attn.out_proj", |
| "model.layers.34.linear_attn.conv1d", |
| "model.layers.34.linear_attn.in_proj_a", |
| "model.layers.34.linear_attn.in_proj_b", |
| "model.layers.34.linear_attn.in_proj_qkv", |
| "model.layers.34.linear_attn.in_proj_z", |
| "model.layers.34.linear_attn.out_proj", |
| "model.layers.35.self_attn.k_proj", |
| "model.layers.35.self_attn.o_proj", |
| "model.layers.35.self_attn.q_proj", |
| "model.layers.35.self_attn.v_proj", |
| "model.layers.36.linear_attn.conv1d", |
| "model.layers.36.linear_attn.in_proj_a", |
| "model.layers.36.linear_attn.in_proj_b", |
| "model.layers.36.linear_attn.in_proj_qkv", |
| "model.layers.36.linear_attn.in_proj_z", |
| "model.layers.36.linear_attn.out_proj", |
| "model.layers.37.linear_attn.conv1d", |
| "model.layers.37.linear_attn.in_proj_a", |
| "model.layers.37.linear_attn.in_proj_b", |
| "model.layers.37.linear_attn.in_proj_qkv", |
| "model.layers.37.linear_attn.in_proj_z", |
| "model.layers.37.linear_attn.out_proj", |
| "model.layers.38.linear_attn.conv1d", |
| "model.layers.38.linear_attn.in_proj_a", |
| "model.layers.38.linear_attn.in_proj_b", |
| "model.layers.38.linear_attn.in_proj_qkv", |
| "model.layers.38.linear_attn.in_proj_z", |
| "model.layers.38.linear_attn.out_proj", |
| "model.layers.39.self_attn.k_proj", |
| "model.layers.39.self_attn.o_proj", |
| "model.layers.39.self_attn.q_proj", |
| "model.layers.39.self_attn.v_proj", |
| "model.layers.4.linear_attn.conv1d", |
| "model.layers.4.linear_attn.in_proj_a", |
| "model.layers.4.linear_attn.in_proj_b", |
| "model.layers.4.linear_attn.in_proj_qkv", |
| "model.layers.4.linear_attn.in_proj_z", |
| "model.layers.4.linear_attn.out_proj", |
| "model.layers.40.linear_attn.conv1d", |
| "model.layers.40.linear_attn.in_proj_a", |
| "model.layers.40.linear_attn.in_proj_b", |
| "model.layers.40.linear_attn.in_proj_qkv", |
| "model.layers.40.linear_attn.in_proj_z", |
| "model.layers.40.linear_attn.out_proj", |
| "model.layers.41.linear_attn.conv1d", |
| "model.layers.41.linear_attn.in_proj_a", |
| "model.layers.41.linear_attn.in_proj_b", |
| "model.layers.41.linear_attn.in_proj_qkv", |
| "model.layers.41.linear_attn.in_proj_z", |
| "model.layers.41.linear_attn.out_proj", |
| "model.layers.42.linear_attn.conv1d", |
| "model.layers.42.linear_attn.in_proj_a", |
| "model.layers.42.linear_attn.in_proj_b", |
| "model.layers.42.linear_attn.in_proj_qkv", |
| "model.layers.42.linear_attn.in_proj_z", |
| "model.layers.42.linear_attn.out_proj", |
| "model.layers.43.self_attn.k_proj", |
| "model.layers.43.self_attn.o_proj", |
| "model.layers.43.self_attn.q_proj", |
| "model.layers.43.self_attn.v_proj", |
| "model.layers.44.linear_attn.conv1d", |
| "model.layers.44.linear_attn.in_proj_a", |
| "model.layers.44.linear_attn.in_proj_b", |
| "model.layers.44.linear_attn.in_proj_qkv", |
| "model.layers.44.linear_attn.in_proj_z", |
| "model.layers.44.linear_attn.out_proj", |
| "model.layers.45.linear_attn.conv1d", |
| "model.layers.45.linear_attn.in_proj_a", |
| "model.layers.45.linear_attn.in_proj_b", |
| "model.layers.45.linear_attn.in_proj_qkv", |
| "model.layers.45.linear_attn.in_proj_z", |
| "model.layers.45.linear_attn.out_proj", |
| "model.layers.46.linear_attn.conv1d", |
| "model.layers.46.linear_attn.in_proj_a", |
| "model.layers.46.linear_attn.in_proj_b", |
| "model.layers.46.linear_attn.in_proj_qkv", |
| "model.layers.46.linear_attn.in_proj_z", |
| "model.layers.46.linear_attn.out_proj", |
| "model.layers.47.self_attn.k_proj", |
| "model.layers.47.self_attn.o_proj", |
| "model.layers.47.self_attn.q_proj", |
| "model.layers.47.self_attn.v_proj", |
| "model.layers.48.linear_attn.conv1d", |
| "model.layers.48.linear_attn.in_proj_a", |
| "model.layers.48.linear_attn.in_proj_b", |
| "model.layers.48.linear_attn.in_proj_qkv", |
| "model.layers.48.linear_attn.in_proj_z", |
| "model.layers.48.linear_attn.out_proj", |
| "model.layers.49.linear_attn.conv1d", |
| "model.layers.49.linear_attn.in_proj_a", |
| "model.layers.49.linear_attn.in_proj_b", |
| "model.layers.49.linear_attn.in_proj_qkv", |
| "model.layers.49.linear_attn.in_proj_z", |
| "model.layers.49.linear_attn.out_proj", |
| "model.layers.5.linear_attn.conv1d", |
| "model.layers.5.linear_attn.in_proj_a", |
| "model.layers.5.linear_attn.in_proj_b", |
| "model.layers.5.linear_attn.in_proj_qkv", |
| "model.layers.5.linear_attn.in_proj_z", |
| "model.layers.5.linear_attn.out_proj", |
| "model.layers.50.linear_attn.conv1d", |
| "model.layers.50.linear_attn.in_proj_a", |
| "model.layers.50.linear_attn.in_proj_b", |
| "model.layers.50.linear_attn.in_proj_qkv", |
| "model.layers.50.linear_attn.in_proj_z", |
| "model.layers.50.linear_attn.out_proj", |
| "model.layers.51.self_attn.k_proj", |
| "model.layers.51.self_attn.o_proj", |
| "model.layers.51.self_attn.q_proj", |
| "model.layers.51.self_attn.v_proj", |
| "model.layers.52.linear_attn.conv1d", |
| "model.layers.52.linear_attn.in_proj_a", |
| "model.layers.52.linear_attn.in_proj_b", |
| "model.layers.52.linear_attn.in_proj_qkv", |
| "model.layers.52.linear_attn.in_proj_z", |
| "model.layers.52.linear_attn.out_proj", |
| "model.layers.53.linear_attn.conv1d", |
| "model.layers.53.linear_attn.in_proj_a", |
| "model.layers.53.linear_attn.in_proj_b", |
| "model.layers.53.linear_attn.in_proj_qkv", |
| "model.layers.53.linear_attn.in_proj_z", |
| "model.layers.53.linear_attn.out_proj", |
| "model.layers.54.linear_attn.conv1d", |
| "model.layers.54.linear_attn.in_proj_a", |
| "model.layers.54.linear_attn.in_proj_b", |
| "model.layers.54.linear_attn.in_proj_qkv", |
| "model.layers.54.linear_attn.in_proj_z", |
| "model.layers.54.linear_attn.out_proj", |
| "model.layers.55.self_attn.k_proj", |
| "model.layers.55.self_attn.o_proj", |
| "model.layers.55.self_attn.q_proj", |
| "model.layers.55.self_attn.v_proj", |
| "model.layers.56.linear_attn.conv1d", |
| "model.layers.56.linear_attn.in_proj_a", |
| "model.layers.56.linear_attn.in_proj_b", |
| "model.layers.56.linear_attn.in_proj_qkv", |
| "model.layers.56.linear_attn.in_proj_z", |
| "model.layers.56.linear_attn.out_proj", |
| "model.layers.57.linear_attn.conv1d", |
| "model.layers.57.linear_attn.in_proj_a", |
| "model.layers.57.linear_attn.in_proj_b", |
| "model.layers.57.linear_attn.in_proj_qkv", |
| "model.layers.57.linear_attn.in_proj_z", |
| "model.layers.57.linear_attn.out_proj", |
| "model.layers.58.linear_attn.conv1d", |
| "model.layers.58.linear_attn.in_proj_a", |
| "model.layers.58.linear_attn.in_proj_b", |
| "model.layers.58.linear_attn.in_proj_qkv", |
| "model.layers.58.linear_attn.in_proj_z", |
| "model.layers.58.linear_attn.out_proj", |
| "model.layers.59.self_attn.k_proj", |
| "model.layers.59.self_attn.o_proj", |
| "model.layers.59.self_attn.q_proj", |
| "model.layers.59.self_attn.v_proj", |
| "model.layers.6.linear_attn.conv1d", |
| "model.layers.6.linear_attn.in_proj_a", |
| "model.layers.6.linear_attn.in_proj_b", |
| "model.layers.6.linear_attn.in_proj_qkv", |
| "model.layers.6.linear_attn.in_proj_z", |
| "model.layers.6.linear_attn.out_proj", |
| "model.layers.60.linear_attn.conv1d", |
| "model.layers.60.linear_attn.in_proj_a", |
| "model.layers.60.linear_attn.in_proj_b", |
| "model.layers.60.linear_attn.in_proj_qkv", |
| "model.layers.60.linear_attn.in_proj_z", |
| "model.layers.60.linear_attn.out_proj", |
| "model.layers.61.linear_attn.conv1d", |
| "model.layers.61.linear_attn.in_proj_a", |
| "model.layers.61.linear_attn.in_proj_b", |
| "model.layers.61.linear_attn.in_proj_qkv", |
| "model.layers.61.linear_attn.in_proj_z", |
| "model.layers.61.linear_attn.out_proj", |
| "model.layers.62.linear_attn.conv1d", |
| "model.layers.62.linear_attn.in_proj_a", |
| "model.layers.62.linear_attn.in_proj_b", |
| "model.layers.62.linear_attn.in_proj_qkv", |
| "model.layers.62.linear_attn.in_proj_z", |
| "model.layers.62.linear_attn.out_proj", |
| "model.layers.63.self_attn.k_proj", |
| "model.layers.63.self_attn.o_proj", |
| "model.layers.63.self_attn.q_proj", |
| "model.layers.63.self_attn.v_proj", |
| "model.layers.64.linear_attn.conv1d", |
| "model.layers.64.linear_attn.in_proj_a", |
| "model.layers.64.linear_attn.in_proj_b", |
| "model.layers.64.linear_attn.in_proj_qkv", |
| "model.layers.64.linear_attn.in_proj_z", |
| "model.layers.64.linear_attn.out_proj", |
| "model.layers.65.linear_attn.conv1d", |
| "model.layers.65.linear_attn.in_proj_a", |
| "model.layers.65.linear_attn.in_proj_b", |
| "model.layers.65.linear_attn.in_proj_qkv", |
| "model.layers.65.linear_attn.in_proj_z", |
| "model.layers.65.linear_attn.out_proj", |
| "model.layers.66.linear_attn.conv1d", |
| "model.layers.66.linear_attn.in_proj_a", |
| "model.layers.66.linear_attn.in_proj_b", |
| "model.layers.66.linear_attn.in_proj_qkv", |
| "model.layers.66.linear_attn.in_proj_z", |
| "model.layers.66.linear_attn.out_proj", |
| "model.layers.67.self_attn.k_proj", |
| "model.layers.67.self_attn.o_proj", |
| "model.layers.67.self_attn.q_proj", |
| "model.layers.67.self_attn.v_proj", |
| "model.layers.68.linear_attn.conv1d", |
| "model.layers.68.linear_attn.in_proj_a", |
| "model.layers.68.linear_attn.in_proj_b", |
| "model.layers.68.linear_attn.in_proj_qkv", |
| "model.layers.68.linear_attn.in_proj_z", |
| "model.layers.68.linear_attn.out_proj", |
| "model.layers.69.linear_attn.conv1d", |
| "model.layers.69.linear_attn.in_proj_a", |
| "model.layers.69.linear_attn.in_proj_b", |
| "model.layers.69.linear_attn.in_proj_qkv", |
| "model.layers.69.linear_attn.in_proj_z", |
| "model.layers.69.linear_attn.out_proj", |
| "model.layers.7.self_attn.k_proj", |
| "model.layers.7.self_attn.o_proj", |
| "model.layers.7.self_attn.q_proj", |
| "model.layers.7.self_attn.v_proj", |
| "model.layers.70.linear_attn.conv1d", |
| "model.layers.70.linear_attn.in_proj_a", |
| "model.layers.70.linear_attn.in_proj_b", |
| "model.layers.70.linear_attn.in_proj_qkv", |
| "model.layers.70.linear_attn.in_proj_z", |
| "model.layers.70.linear_attn.out_proj", |
| "model.layers.71.self_attn.k_proj", |
| "model.layers.71.self_attn.o_proj", |
| "model.layers.71.self_attn.q_proj", |
| "model.layers.71.self_attn.v_proj", |
| "model.layers.72.linear_attn.conv1d", |
| "model.layers.72.linear_attn.in_proj_a", |
| "model.layers.72.linear_attn.in_proj_b", |
| "model.layers.72.linear_attn.in_proj_qkv", |
| "model.layers.72.linear_attn.in_proj_z", |
| "model.layers.72.linear_attn.out_proj", |
| "model.layers.73.linear_attn.conv1d", |
| "model.layers.73.linear_attn.in_proj_a", |
| "model.layers.73.linear_attn.in_proj_b", |
| "model.layers.73.linear_attn.in_proj_qkv", |
| "model.layers.73.linear_attn.in_proj_z", |
| "model.layers.73.linear_attn.out_proj", |
| "model.layers.74.linear_attn.conv1d", |
| "model.layers.74.linear_attn.in_proj_a", |
| "model.layers.74.linear_attn.in_proj_b", |
| "model.layers.74.linear_attn.in_proj_qkv", |
| "model.layers.74.linear_attn.in_proj_z", |
| "model.layers.74.linear_attn.out_proj", |
| "model.layers.75.self_attn.k_proj", |
| "model.layers.75.self_attn.o_proj", |
| "model.layers.75.self_attn.q_proj", |
| "model.layers.75.self_attn.v_proj", |
| "model.layers.76.linear_attn.conv1d", |
| "model.layers.76.linear_attn.in_proj_a", |
| "model.layers.76.linear_attn.in_proj_b", |
| "model.layers.76.linear_attn.in_proj_qkv", |
| "model.layers.76.linear_attn.in_proj_z", |
| "model.layers.76.linear_attn.out_proj", |
| "model.layers.77.linear_attn.conv1d", |
| "model.layers.77.linear_attn.in_proj_a", |
| "model.layers.77.linear_attn.in_proj_b", |
| "model.layers.77.linear_attn.in_proj_qkv", |
| "model.layers.77.linear_attn.in_proj_z", |
| "model.layers.77.linear_attn.out_proj", |
| "model.layers.78.linear_attn.conv1d", |
| "model.layers.78.linear_attn.in_proj_a", |
| "model.layers.78.linear_attn.in_proj_b", |
| "model.layers.78.linear_attn.in_proj_qkv", |
| "model.layers.78.linear_attn.in_proj_z", |
| "model.layers.78.linear_attn.out_proj", |
| "model.layers.79.self_attn.k_proj", |
| "model.layers.79.self_attn.o_proj", |
| "model.layers.79.self_attn.q_proj", |
| "model.layers.79.self_attn.v_proj", |
| "model.layers.8.linear_attn.conv1d", |
| "model.layers.8.linear_attn.in_proj_a", |
| "model.layers.8.linear_attn.in_proj_b", |
| "model.layers.8.linear_attn.in_proj_qkv", |
| "model.layers.8.linear_attn.in_proj_z", |
| "model.layers.8.linear_attn.out_proj", |
| "model.layers.80.linear_attn.conv1d", |
| "model.layers.80.linear_attn.in_proj_a", |
| "model.layers.80.linear_attn.in_proj_b", |
| "model.layers.80.linear_attn.in_proj_qkv", |
| "model.layers.80.linear_attn.in_proj_z", |
| "model.layers.80.linear_attn.out_proj", |
| "model.layers.81.linear_attn.conv1d", |
| "model.layers.81.linear_attn.in_proj_a", |
| "model.layers.81.linear_attn.in_proj_b", |
| "model.layers.81.linear_attn.in_proj_qkv", |
| "model.layers.81.linear_attn.in_proj_z", |
| "model.layers.81.linear_attn.out_proj", |
| "model.layers.82.linear_attn.conv1d", |
| "model.layers.82.linear_attn.in_proj_a", |
| "model.layers.82.linear_attn.in_proj_b", |
| "model.layers.82.linear_attn.in_proj_qkv", |
| "model.layers.82.linear_attn.in_proj_z", |
| "model.layers.82.linear_attn.out_proj", |
| "model.layers.83.self_attn.k_proj", |
| "model.layers.83.self_attn.o_proj", |
| "model.layers.83.self_attn.q_proj", |
| "model.layers.83.self_attn.v_proj", |
| "model.layers.84.linear_attn.conv1d", |
| "model.layers.84.linear_attn.in_proj_a", |
| "model.layers.84.linear_attn.in_proj_b", |
| "model.layers.84.linear_attn.in_proj_qkv", |
| "model.layers.84.linear_attn.in_proj_z", |
| "model.layers.84.linear_attn.out_proj", |
| "model.layers.85.linear_attn.conv1d", |
| "model.layers.85.linear_attn.in_proj_a", |
| "model.layers.85.linear_attn.in_proj_b", |
| "model.layers.85.linear_attn.in_proj_qkv", |
| "model.layers.85.linear_attn.in_proj_z", |
| "model.layers.85.linear_attn.out_proj", |
| "model.layers.86.linear_attn.conv1d", |
| "model.layers.86.linear_attn.in_proj_a", |
| "model.layers.86.linear_attn.in_proj_b", |
| "model.layers.86.linear_attn.in_proj_qkv", |
| "model.layers.86.linear_attn.in_proj_z", |
| "model.layers.86.linear_attn.out_proj", |
| "model.layers.87.self_attn.k_proj", |
| "model.layers.87.self_attn.o_proj", |
| "model.layers.87.self_attn.q_proj", |
| "model.layers.87.self_attn.v_proj", |
| "model.layers.88.linear_attn.conv1d", |
| "model.layers.88.linear_attn.in_proj_a", |
| "model.layers.88.linear_attn.in_proj_b", |
| "model.layers.88.linear_attn.in_proj_qkv", |
| "model.layers.88.linear_attn.in_proj_z", |
| "model.layers.88.linear_attn.out_proj", |
| "model.layers.89.linear_attn.conv1d", |
| "model.layers.89.linear_attn.in_proj_a", |
| "model.layers.89.linear_attn.in_proj_b", |
| "model.layers.89.linear_attn.in_proj_qkv", |
| "model.layers.89.linear_attn.in_proj_z", |
| "model.layers.89.linear_attn.out_proj", |
| "model.layers.9.linear_attn.conv1d", |
| "model.layers.9.linear_attn.in_proj_a", |
| "model.layers.9.linear_attn.in_proj_b", |
| "model.layers.9.linear_attn.in_proj_qkv", |
| "model.layers.9.linear_attn.in_proj_z", |
| "model.layers.9.linear_attn.out_proj", |
| "model.layers.90.linear_attn.conv1d", |
| "model.layers.90.linear_attn.in_proj_a", |
| "model.layers.90.linear_attn.in_proj_b", |
| "model.layers.90.linear_attn.in_proj_qkv", |
| "model.layers.90.linear_attn.in_proj_z", |
| "model.layers.90.linear_attn.out_proj", |
| "model.layers.91.self_attn.k_proj", |
| "model.layers.91.self_attn.o_proj", |
| "model.layers.91.self_attn.q_proj", |
| "model.layers.91.self_attn.v_proj" |
| ] |
| }, |
| "group_1": { |
| "input_activations": { |
| "dynamic": false, |
| "num_bits": 4, |
| "type": "float", |
| "group_size": 16 |
| }, |
| "weights": { |
| "dynamic": false, |
| "num_bits": 4, |
| "type": "float", |
| "group_size": 16 |
| }, |
| "targets": [ |
| "model.layers.0.mlp.experts", |
| "model.layers.1.mlp.experts", |
| "model.layers.10.mlp.experts", |
| "model.layers.11.mlp.experts", |
| "model.layers.12.mlp.experts", |
| "model.layers.13.mlp.experts", |
| "model.layers.14.mlp.experts", |
| "model.layers.15.mlp.experts", |
| "model.layers.16.mlp.experts", |
| "model.layers.17.mlp.experts", |
| "model.layers.18.mlp.experts", |
| "model.layers.19.mlp.experts", |
| "model.layers.2.mlp.experts", |
| "model.layers.20.mlp.experts", |
| "model.layers.21.mlp.experts", |
| "model.layers.22.mlp.experts", |
| "model.layers.23.mlp.experts", |
| "model.layers.24.mlp.experts", |
| "model.layers.25.mlp.experts", |
| "model.layers.26.mlp.experts", |
| "model.layers.27.mlp.experts", |
| "model.layers.28.mlp.experts", |
| "model.layers.29.mlp.experts", |
| "model.layers.3.mlp.experts", |
| "model.layers.30.mlp.experts", |
| "model.layers.31.mlp.experts", |
| "model.layers.32.mlp.experts", |
| "model.layers.33.mlp.experts", |
| "model.layers.34.mlp.experts", |
| "model.layers.35.mlp.experts", |
| "model.layers.36.mlp.experts", |
| "model.layers.37.mlp.experts", |
| "model.layers.38.mlp.experts", |
| "model.layers.39.mlp.experts", |
| "model.layers.4.mlp.experts", |
| "model.layers.40.mlp.experts", |
| "model.layers.41.mlp.experts", |
| "model.layers.42.mlp.experts", |
| "model.layers.43.mlp.experts", |
| "model.layers.44.mlp.experts", |
| "model.layers.45.mlp.experts", |
| "model.layers.46.mlp.experts", |
| "model.layers.47.mlp.experts", |
| "model.layers.48.mlp.experts", |
| "model.layers.49.mlp.experts", |
| "model.layers.5.mlp.experts", |
| "model.layers.50.mlp.experts", |
| "model.layers.51.mlp.experts", |
| "model.layers.52.mlp.experts", |
| "model.layers.53.mlp.experts", |
| "model.layers.54.mlp.experts", |
| "model.layers.55.mlp.experts", |
| "model.layers.56.mlp.experts", |
| "model.layers.57.mlp.experts", |
| "model.layers.58.mlp.experts", |
| "model.layers.59.mlp.experts", |
| "model.layers.6.mlp.experts", |
| "model.layers.60.mlp.experts", |
| "model.layers.61.mlp.experts", |
| "model.layers.62.mlp.experts", |
| "model.layers.63.mlp.experts", |
| "model.layers.64.mlp.experts", |
| "model.layers.65.mlp.experts", |
| "model.layers.66.mlp.experts", |
| "model.layers.67.mlp.experts", |
| "model.layers.68.mlp.experts", |
| "model.layers.69.mlp.experts", |
| "model.layers.7.mlp.experts", |
| "model.layers.70.mlp.experts", |
| "model.layers.71.mlp.experts", |
| "model.layers.72.mlp.experts", |
| "model.layers.73.mlp.experts", |
| "model.layers.74.mlp.experts", |
| "model.layers.75.mlp.experts", |
| "model.layers.76.mlp.experts", |
| "model.layers.77.mlp.experts", |
| "model.layers.78.mlp.experts", |
| "model.layers.79.mlp.experts", |
| "model.layers.8.mlp.experts", |
| "model.layers.80.mlp.experts", |
| "model.layers.81.mlp.experts", |
| "model.layers.82.mlp.experts", |
| "model.layers.83.mlp.experts", |
| "model.layers.84.mlp.experts", |
| "model.layers.85.mlp.experts", |
| "model.layers.86.mlp.experts", |
| "model.layers.87.mlp.experts", |
| "model.layers.88.mlp.experts", |
| "model.layers.89.mlp.experts", |
| "model.layers.9.mlp.experts", |
| "model.layers.90.mlp.experts", |
| "model.layers.91.mlp.experts" |
| ] |
| } |
| }, |
| "quantized_layers": { |
| "model.layers.0.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.0.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.0.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.0.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.0.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.0.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.0.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.1.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.1.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.1.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.1.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.1.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.1.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.1.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.2.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.2.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.2.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.2.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.2.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.2.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.2.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.3.self_attn.q_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.3.self_attn.k_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.3.self_attn.v_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.3.self_attn.o_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.3.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.4.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.4.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.4.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.4.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.4.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.4.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.4.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.5.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.5.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.5.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.5.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.5.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.5.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.5.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.6.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.6.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.6.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.6.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.6.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.6.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.6.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.7.self_attn.q_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.7.self_attn.k_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.7.self_attn.v_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.7.self_attn.o_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.7.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.8.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.8.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.8.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.8.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.8.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.8.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.8.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.9.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.9.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.9.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.9.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.9.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.9.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.9.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.10.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.10.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.10.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.10.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.10.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.10.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.10.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.11.self_attn.q_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.11.self_attn.k_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.11.self_attn.v_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.11.self_attn.o_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.11.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.12.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.12.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.12.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.12.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.12.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.12.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.12.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.13.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.13.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.13.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.13.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.13.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.13.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.13.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.14.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.14.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.14.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.14.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.14.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.14.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.14.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.15.self_attn.q_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.15.self_attn.k_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.15.self_attn.v_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.15.self_attn.o_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.15.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.16.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.16.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.16.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.16.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.16.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.16.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.16.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.17.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.17.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.17.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.17.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.17.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.17.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.17.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.18.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.18.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.18.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.18.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.18.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.18.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.18.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.19.self_attn.q_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.19.self_attn.k_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.19.self_attn.v_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.19.self_attn.o_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.19.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.20.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.20.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.20.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.20.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.20.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.20.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.20.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.21.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.21.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.21.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.21.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.21.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.21.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.21.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.22.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.22.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.22.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.22.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.22.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.22.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.22.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.23.self_attn.q_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.23.self_attn.k_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.23.self_attn.v_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.23.self_attn.o_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.23.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.24.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.24.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.24.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.24.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.24.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.24.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.24.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.25.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.25.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.25.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.25.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.25.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.25.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.25.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.26.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.26.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.26.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.26.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.26.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.26.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.26.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.27.self_attn.q_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.27.self_attn.k_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.27.self_attn.v_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.27.self_attn.o_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.27.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.28.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.28.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.28.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.28.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.28.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.28.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.28.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.29.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.29.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.29.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.29.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.29.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.29.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.29.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.30.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.30.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.30.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.30.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.30.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.30.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.30.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.31.self_attn.q_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.31.self_attn.k_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.31.self_attn.v_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.31.self_attn.o_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.31.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.32.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.32.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.32.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.32.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.32.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.32.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.32.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.33.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.33.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.33.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.33.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.33.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.33.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.33.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.34.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.34.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.34.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.34.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.34.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.34.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.34.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.35.self_attn.q_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.35.self_attn.k_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.35.self_attn.v_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.35.self_attn.o_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.35.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.36.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.36.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.36.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.36.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.36.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.36.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.36.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.37.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.37.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.37.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.37.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.37.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.37.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.37.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.38.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.38.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.38.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.38.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.38.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.38.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.38.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.39.self_attn.q_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.39.self_attn.k_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.39.self_attn.v_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.39.self_attn.o_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.39.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.40.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.40.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.40.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.40.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.40.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.40.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.40.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.41.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.41.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.41.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.41.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.41.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.41.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.41.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.42.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.42.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.42.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.42.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.42.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.42.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.42.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.43.self_attn.q_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.43.self_attn.k_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.43.self_attn.v_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.43.self_attn.o_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.43.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.44.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.44.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.44.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.44.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.44.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.44.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.44.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.45.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.45.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.45.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.45.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.45.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.45.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.45.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.46.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.46.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.46.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.46.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.46.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.46.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.46.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.47.self_attn.q_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.47.self_attn.k_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.47.self_attn.v_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.47.self_attn.o_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.47.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.48.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.48.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.48.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.48.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.48.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.48.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.48.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.49.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.49.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.49.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.49.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.49.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.49.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.49.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.50.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.50.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.50.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.50.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.50.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.50.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.50.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.51.self_attn.q_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.51.self_attn.k_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.51.self_attn.v_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.51.self_attn.o_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.51.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.52.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.52.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.52.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.52.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.52.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.52.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.52.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.53.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.53.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.53.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.53.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.53.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.53.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.53.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.54.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.54.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.54.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.54.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.54.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.54.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.54.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.55.self_attn.q_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.55.self_attn.k_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.55.self_attn.v_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.55.self_attn.o_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.55.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.56.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.56.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.56.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.56.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.56.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.56.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.56.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.57.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.57.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.57.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.57.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.57.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.57.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.57.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.58.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.58.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.58.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.58.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.58.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.58.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.58.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.59.self_attn.q_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.59.self_attn.k_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.59.self_attn.v_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.59.self_attn.o_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.59.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.60.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.60.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.60.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.60.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.60.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.60.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.60.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.61.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.61.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.61.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.61.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.61.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.61.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.61.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.62.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.62.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.62.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.62.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.62.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.62.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.62.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.63.self_attn.q_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.63.self_attn.k_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.63.self_attn.v_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.63.self_attn.o_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.63.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.64.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.64.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.64.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.64.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.64.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.64.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.64.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.65.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.65.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.65.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.65.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.65.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.65.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.65.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.66.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.66.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.66.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.66.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.66.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.66.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.66.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.67.self_attn.q_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.67.self_attn.k_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.67.self_attn.v_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.67.self_attn.o_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.67.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.68.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.68.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.68.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.68.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.68.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.68.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.68.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.69.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.69.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.69.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.69.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.69.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.69.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.69.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.70.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.70.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.70.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.70.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.70.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.70.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.70.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.71.self_attn.q_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.71.self_attn.k_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.71.self_attn.v_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.71.self_attn.o_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.71.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.72.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.72.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.72.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.72.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.72.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.72.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.72.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.73.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.73.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.73.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.73.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.73.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.73.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.73.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.74.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.74.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.74.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.74.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.74.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.74.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.74.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.75.self_attn.q_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.75.self_attn.k_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.75.self_attn.v_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.75.self_attn.o_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.75.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.76.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.76.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.76.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.76.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.76.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.76.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.76.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.77.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.77.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.77.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.77.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.77.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.77.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.77.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.78.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.78.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.78.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.78.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.78.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.78.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.78.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.79.self_attn.q_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.79.self_attn.k_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.79.self_attn.v_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.79.self_attn.o_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.79.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.80.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.80.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.80.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.80.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.80.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.80.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.80.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.81.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.81.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.81.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.81.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.81.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.81.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.81.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.82.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.82.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.82.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.82.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.82.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.82.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.82.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.83.self_attn.q_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.83.self_attn.k_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.83.self_attn.v_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.83.self_attn.o_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.83.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.84.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.84.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.84.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.84.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.84.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.84.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.84.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.85.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.85.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.85.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.85.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.85.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.85.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.85.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.86.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.86.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.86.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.86.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.86.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.86.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.86.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.87.self_attn.q_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.87.self_attn.k_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.87.self_attn.v_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.87.self_attn.o_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.87.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.88.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.88.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.88.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.88.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.88.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.88.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.88.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.89.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.89.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.89.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.89.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.89.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.89.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.89.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.90.linear_attn.conv1d": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.90.linear_attn.out_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.90.linear_attn.in_proj_qkv": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.90.linear_attn.in_proj_z": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.90.linear_attn.in_proj_b": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.90.linear_attn.in_proj_a": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.90.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| }, |
| "model.layers.91.self_attn.q_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.91.self_attn.k_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.91.self_attn.v_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.91.self_attn.o_proj": { |
| "quant_algo": "FP8" |
| }, |
| "model.layers.91.mlp.experts": { |
| "quant_algo": "NVFP4", |
| "group_size": 16 |
| } |
| }, |
| "ignore": [ |
| "mtp*", |
| "mtp.layers.0*" |
| ], |
| "quant_algo": "MIXED_PRECISION", |
| "kv_cache_scheme": { |
| "dynamic": false, |
| "num_bits": 8, |
| "type": "float" |
| }, |
| "producer": { |
| "name": "modelopt", |
| "version": "0.42.0" |
| }, |
| "quant_method": "modelopt" |
| } |
| } |
