| quant_stage: | |
| quant_modifiers: | |
| AWQModifier: | |
| mappings: | |
| - smooth_layer: re:.*input_layernorm$ | |
| balance_layers: ['re:.*q_proj$', 're:.*k_proj$', 're:.*v_proj$'] | |
| activation_hook_target: null | |
| - smooth_layer: re:.*v_proj$ | |
| balance_layers: ['re:.*o_proj$'] | |
| activation_hook_target: null | |
| - smooth_layer: re:.*post_attention_layernorm$ | |
| balance_layers: ['re:.*gate_proj$', 're:.*up_proj$'] | |
| activation_hook_target: null | |
| - smooth_layer: re:.*up_proj$ | |
| balance_layers: ['re:.*down_proj$'] | |
| activation_hook_target: null | |
| duo_scaling: both | |
| n_grid: 20 | |
| QuantizationModifier: | |
| config_groups: | |
| group_0: | |
| targets: [Linear] | |
| weights: | |
| num_bits: 4 | |
| type: int | |
| symmetric: true | |
| group_size: 128 | |
| strategy: group | |
| block_structure: null | |
| dynamic: false | |
| actorder: null | |
| scale_dtype: null | |
| zp_dtype: null | |
| observer: memoryless_minmax | |
| observer_kwargs: {} | |
| input_activations: null | |
| output_activations: null | |
| format: null | |
| targets: [Linear] | |
| ignore: [lm_head] | |
| bypass_divisibility_checks: false | |