neural-compressor icon indicating copy to clipboard operation
neural-compressor copied to clipboard

Support save/load API for WOQ

Open Kaihui-intel opened this issue 1 year ago • 0 comments

Type of Change

feature

Description

Support save/load API for WOQ remove export_compressed_model from config

Expected Behavior & Potential Risk

the expected behavior that triggered by this PR

How has this PR been tested?

UT local test: fp32&rtn

ks Version Filter n-shot Metric Value Stderr
lambada_openai 1 none 0 perplexity 26.0209 ± 0.9382
none 0 acc 0.3790 ± 0.0068

Accuracy: 0.37900 Batch size = 1

Tasks Version Filter n-shot Metric Value Stderr
lambada_openai 1 none 0 perplexity 29.1191 ± 1.1134
none 0 acc 0.3679 ± 0.0067

Accuracy: 0.36794 Batch size = 1

opt_125m_woq_gptq_int4_dq_bnb

ks Version Filter n-shot Metric Value Stderr
lambada_openai 1 none 0 perplexity 26.9172 ± 1.0165
none 0 acc 0.3701 ± 0.0067
Accuracy: 0.37008
Batch size = 1

Dependency Change?

any library dependency introduced or removed

Kaihui-intel avatar May 11 '24 01:05 Kaihui-intel