| default_stage: | |
| default_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 | |
| scale_dtype: null | |
| zp_dtype: null | |
| observer: mse | |
| observer_kwargs: {} | |
| input_activations: null | |
| output_activations: null | |
| format: null | |
| targets: [Linear] | |
| ignore: [model.embed_tokens, 're:.*linear_attn.*', 're:.*norm.*', 're:.*shared_expert.*', | |
| 're:.*shared_expert_gate$', 're:.*mlp[.]gate$', 're:.*router.*', 're:.*self_attn.*', | |
| lm_head] | |
| bypass_divisibility_checks: false | |
| mappings: | |
| - smooth_layer: re:.*post_attention_layernorm | |
| balance_layers: ['re:.*mlp.gate', 're:.*mlp.shared_expert.gate_proj', 're:.*mlp.shared_expert.up_proj', | |
| 're:.*mlp.shared_expert_gate', 're:.*mlp.experts.*gate_proj', 're:.*mlp.experts.*up_proj'] | |
| activation_hook_target: null | |
| balance_exponent: 1 | |
| - smooth_layer: re:.*mlp.experts.*up_proj | |
| balance_layers: ['re:.*mlp.experts.*down_proj'] | |
| activation_hook_target: null | |
| balance_exponent: 1 | |
| offload_device: !!python/object/apply:torch.device [cpu] | |
| duo_scaling: true | |
| n_grid: 20 | |