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
This commit is contained in:
16
recipe.yaml
Normal file
16
recipe.yaml
Normal file
@@ -0,0 +1,16 @@
|
||||
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
|
||||
Reference in New Issue
Block a user