{ "quant_method": "smoothquant", "method_name": "SmoothQuant", "format": "w8a8_int8", "w_bits": 8, "a_bits": 8, "weight_quant": "per_channel", "act_quant": "per_token", "act_dynamic": true, "w_sym": true, "a_sym": true, "quantize_bmm_input": false, "weight_dtype": "int8", "weight_scale_dtype": "float16", "weight_scale_shape": "[out_features, 1]", "dequant": "weight_fp16 = weight.int8 * weight_scale", "kept_fp16": [ "input_layernorm / post_attention_layernorm (smoothed)", "model.norm", "model.embed_tokens", "lm_head (untied, stored)" ], "smoothing_folded_offline": true, "smoothing_sites": [ "input_layernorm -> self_attn.qkv_proj", "post_attention_layernorm -> mlp.gate_up_proj" ], "runtime_transforms": [], "repo_url": "https://github.com/mit-han-lab/smoothquant", "repo_commit": "c61476d728e42ae0d8a35e7e78494edcac3237b5", "inference_adapter": "quant_agent_inference_adapter.py", "alpha": 0.75, "calibration": { "dataset": "mit-han-lab/pile-val-backup val.jsonl (local copy)", "shuffle_seed": 42, "num_samples": 512, "seq_len": 512 }, "base_model": "microsoft/Phi-3.5-mini-instruct", "quantized_modules": [ "model.layers.0.mlp.down_proj", "model.layers.0.mlp.gate_up_proj", "model.layers.0.self_attn.o_proj", "model.layers.0.self_attn.qkv_proj", "model.layers.1.mlp.down_proj", "model.layers.1.mlp.gate_up_proj", "model.layers.1.self_attn.o_proj", "model.layers.1.self_attn.qkv_proj", "model.layers.10.mlp.down_proj", "model.layers.10.mlp.gate_up_proj", "model.layers.10.self_attn.o_proj", "model.layers.10.self_attn.qkv_proj", "model.layers.11.mlp.down_proj", "model.layers.11.mlp.gate_up_proj", "model.layers.11.self_attn.o_proj", "model.layers.11.self_attn.qkv_proj", "model.layers.12.mlp.down_proj", "model.layers.12.mlp.gate_up_proj", "model.layers.12.self_attn.o_proj", "model.layers.12.self_attn.qkv_proj", "model.layers.13.mlp.down_proj", "model.layers.13.mlp.gate_up_proj", "model.layers.13.self_attn.o_proj", "model.layers.13.self_attn.qkv_proj", "model.layers.14.mlp.down_proj", "model.layers.14.mlp.gate_up_proj", "model.layers.14.self_attn.o_proj", "model.layers.14.self_attn.qkv_proj", "model.layers.15.mlp.down_proj", "model.layers.15.mlp.gate_up_proj", "model.layers.15.self_attn.o_proj", "model.layers.15.self_attn.qkv_proj", "model.layers.16.mlp.down_proj", "model.layers.16.mlp.gate_up_proj", "model.layers.16.self_attn.o_proj", "model.layers.16.self_attn.qkv_proj", "model.layers.17.mlp.down_proj", "model.layers.17.mlp.gate_up_proj", "model.layers.17.self_attn.o_proj", "model.layers.17.self_attn.qkv_proj", "model.layers.18.mlp.down_proj", "model.layers.18.mlp.gate_up_proj", "model.layers.18.self_attn.o_proj", "model.layers.18.self_attn.qkv_proj", "model.layers.19.mlp.down_proj", "model.layers.19.mlp.gate_up_proj", "model.layers.19.self_attn.o_proj", "model.layers.19.self_attn.qkv_proj", "model.layers.2.mlp.down_proj", "model.layers.2.mlp.gate_up_proj", "model.layers.2.self_attn.o_proj", "model.layers.2.self_attn.qkv_proj", "model.layers.20.mlp.down_proj", "model.layers.20.mlp.gate_up_proj", "model.layers.20.self_attn.o_proj", "model.layers.20.self_attn.qkv_proj", "model.layers.21.mlp.down_proj", "model.layers.21.mlp.gate_up_proj", "model.layers.21.self_attn.o_proj", "model.layers.21.self_attn.qkv_proj", "model.layers.22.mlp.down_proj", "model.layers.22.mlp.gate_up_proj", "model.layers.22.self_attn.o_proj", "model.layers.22.self_attn.qkv_proj", "model.layers.23.mlp.down_proj", "model.layers.23.mlp.gate_up_proj", "model.layers.23.self_attn.o_proj", "model.layers.23.self_attn.qkv_proj", "model.layers.24.mlp.down_proj", "model.layers.24.mlp.gate_up_proj", "model.layers.24.self_attn.o_proj", "model.layers.24.self_attn.qkv_proj", "model.layers.25.mlp.down_proj", "model.layers.25.mlp.gate_up_proj", "model.layers.25.self_attn.o_proj", "model.layers.25.self_attn.qkv_proj", "model.layers.26.mlp.down_proj", "model.layers.26.mlp.gate_up_proj", "model.layers.26.self_attn.o_proj", "model.layers.26.self_attn.qkv_proj", "model.layers.27.mlp.down_proj", "model.layers.27.mlp.gate_up_proj", "model.layers.27.self_attn.o_proj", "model.layers.27.self_attn.qkv_proj", "model.layers.28.mlp.down_proj", "model.layers.28.mlp.gate_up_proj", "model.layers.28.self_attn.o_proj", "model.layers.28.self_attn.qkv_proj", "model.layers.29.mlp.down_proj", "model.layers.29.mlp.gate_up_proj", "model.layers.29.self_attn.o_proj", "model.layers.29.self_attn.qkv_proj", "model.layers.3.mlp.down_proj", "model.layers.3.mlp.gate_up_proj", "model.layers.3.self_attn.o_proj", "model.layers.3.self_attn.qkv_proj", "model.layers.30.mlp.down_proj", "model.layers.30.mlp.gate_up_proj", "model.layers.30.self_attn.o_proj", "model.layers.30.self_attn.qkv_proj", "model.layers.31.mlp.down_proj", "model.layers.31.mlp.gate_up_proj", "model.layers.31.self_attn.o_proj", "model.layers.31.self_attn.qkv_proj", "model.layers.4.mlp.down_proj", "model.layers.4.mlp.gate_up_proj", "model.layers.4.self_attn.o_proj", "model.layers.4.self_attn.qkv_proj", "model.layers.5.mlp.down_proj", "model.layers.5.mlp.gate_up_proj", "model.layers.5.self_attn.o_proj", "model.layers.5.self_attn.qkv_proj", "model.layers.6.mlp.down_proj", "model.layers.6.mlp.gate_up_proj", "model.layers.6.self_attn.o_proj", "model.layers.6.self_attn.qkv_proj", "model.layers.7.mlp.down_proj", "model.layers.7.mlp.gate_up_proj", "model.layers.7.self_attn.o_proj", "model.layers.7.self_attn.qkv_proj", "model.layers.8.mlp.down_proj", "model.layers.8.mlp.gate_up_proj", "model.layers.8.self_attn.o_proj", "model.layers.8.self_attn.qkv_proj", "model.layers.9.mlp.down_proj", "model.layers.9.mlp.gate_up_proj", "model.layers.9.self_attn.o_proj", "model.layers.9.self_attn.qkv_proj" ], "num_quantized_linears": 128 }