Files
Qwen3-8B.w8a8/recipe.yaml

9 lines
197 B
YAML
Raw Permalink Normal View History

default_stage:
default_modifiers:
SmoothQuantModifier: {smoothing_strength: 0.7}
GPTQModifier:
dampening_frac: 0.01
scheme: W8A8
targets: Linear
ignore: [lm_head]