default_stage: default_modifiers: AWQModifier: config_groups: group_0: targets: [Linear] weights: num_bits: 8 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: ['re:.*embed_tokens', 're:.*linear_attn.*', 're:model[.]visual.*', 're:mtp.*', lm_head] bypass_divisibility_checks: false mappings: - smooth_layer: re:model.*layers[.](3|7|11|15|19|23|27|31|35|39|43|47|51|55|59|63)[.]input_layernorm balance_layers: ['re:model.*layers[.](3|7|11|15|19|23|27|31|35|39|43|47|51|55|59|63)[.]self_attn[.]q_proj', 're:model.*layers[.](3|7|11|15|19|23|27|31|35|39|43|47|51|55|59|63)[.]self_attn[.]k_proj', 're:model.*layers[.](3|7|11|15|19|23|27|31|35|39|43|47|51|55|59|63)[.]self_attn[.]v_proj'] activation_hook_target: null balance_exponent: 1 - smooth_layer: re:model.*layers[.](3|7|11|15|19|23|27|31|35|39|43|47|51|55|59|63)[.]self_attn[.]v_proj balance_layers: ['re:model.*layers[.](3|7|11|15|19|23|27|31|35|39|43|47|51|55|59|63)[.]self_attn[.]o_proj'] activation_hook_target: null balance_exponent: 1 - smooth_layer: re:model.*post_attention_layernorm balance_layers: ['re:model.*mlp[.]gate_proj', 're:model.*mlp[.]up_proj'] activation_hook_target: null balance_exponent: 1 - smooth_layer: re:model.*mlp[.]up_proj balance_layers: ['re:model.*mlp[.]down_proj'] activation_hook_target: null balance_exponent: 1 duo_scaling: true n_grid: 20