Files
ModelHub XC 1f768d8698 初始化项目,由ModelHub XC社区提供模型
Model: RedHatAI/Phi-3-medium-128k-instruct-quantized.w8a8
Source: Original Platform
2026-09-10 08:16:12 +08:00

17 lines
394 B
YAML

quant_stage:
quant_modifiers:
SmoothQuantModifier:
smoothing_strength: 0.8
mappings:
- - ['re:.*qkv_proj']
- re:.*input_layernorm
- - ['re:.*gate_up_proj']
- re:.*post_attention_layernorm
GPTQModifier:
sequential_update: true
dampening_frac: 0.01
ignore: [lm_head]
scheme: W8A8
targets: Linear
observer: mse