Files
Qwen2.5-32B-quantized.w8a16/recipe.yaml

10 lines
192 B
YAML
Raw Normal View History

quant_stage:
quant_modifiers:
GPTQModifier:
sequential_update: true
dampening_frac: 0.1
ignore: [lm_head]
scheme: W8A16
targets: Linear
observer: mse