Files
Qwen2.5-1.5B-quantized.w4a16/recipe.yaml

12 lines
335 B
YAML
Raw Normal View History

quant_stage:
quant_modifiers:
GPTQModifier:
sequential_update: true
dampening_frac: 0.1
ignore: [lm_head]
config_groups:
group_0:
weights: {num_bits: 4, type: int, symmetric: true, strategy: group, group_size: 128,
actorder: group, observer: mse}
targets: [Linear]