DEFAULT_stage: | |
DEFAULT_modifiers: | |
SmoothQuantModifier: | |
smoothing_strength: 0.8625 | |
mappings: | |
- - ['re:.*qkv_proj'] | |
- re:.*input_layernorm | |
- - ['re:.*gate_up_proj'] | |
- re:.*post_attention_layernorm | |
GPTQModifier: | |
sequential_update: false | |
targets: Linear | |
ignore: [lm_head] | |
scheme: W8A8 | |