| default_stage: | |
| default_modifiers: | |
| AWQModifier: | |
| targets: [Linear] | |
| ignore: [lm_head, model.embed_tokens, 're:.*input_layernorm$', 're:.*post_attention_layernorm$', | |
| model.norm] | |
| scheme: W4A16 | |
| mappings: | |
| - smooth_layer: re:.*input_layernorm$ | |
| balance_layers: ['re:.*self_attn.q_proj$', 're:.*self_attn.k_proj$', 're:.*self_attn.v_proj$', | |
| 're:.*mlp.gate_proj$', 're:.*mlp.up_proj$'] | |
| - smooth_layer: re:.*v_proj$ | |
| balance_layers: ['re:.*o_proj$'] | |
| - smooth_layer: re:.*up_proj$ | |
| balance_layers: ['re:.*down_proj$'] | |
| offload_device: !!python/object/apply:torch.device [cpu] | |
| duo_scaling: true | |