| quant_stage: | |
| quant_modifiers: | |
| AWQModifier: | |
| config_groups: | |
| group_0: | |
| targets: [Linear] | |
| weights: | |
| num_bits: 4 | |
| type: int | |
| symmetric: true | |
| group_size: 32 | |
| strategy: group | |
| block_structure: null | |
| dynamic: false | |
| actorder: null | |
| observer: mse | |
| observer_kwargs: {} | |
| input_activations: null | |
| output_activations: null | |
| format: null | |
| targets: [Linear] | |
| ignore: [word_embeddings, 're:model.layers.0[.].*', 're:.*input_layernorm', 're:.*post_attention_layernorm', | |
| 're:.*layernorm.*', 're:.*attention[.]dense', 're:.*shared_experts.*', 're:.*mlp[.]gate.*', | |
| model.norm, lm_head] | |
| mappings: | |
| - smooth_layer: re:.*input_layernorm | |
| balance_layers: ['re:.*query_key_value'] | |
| - smooth_layer: re:.*post_attention_layernorm | |
| balance_layers: ['re:.*gate_proj', 're:.*up_proj'] | |
| - smooth_layer: re:.*mlp[.]experts[.]\d+[.]up_proj$ | |
| balance_layers: ['re:.*mlp[.]experts[.]\d+[.]down_proj$'] | |
| duo_scaling: true | |