Skip to content

Latest commit

 

History

History
57 lines (40 loc) · 1.26 KB

post_process.md

File metadata and controls

57 lines (40 loc) · 1.26 KB

Post-Process

Post-quantization process classes for improving quantized model accuracy.

Base Class

::: onecomp.post_process.PostQuantizationProcess options: show_source: false

Global PTQ

::: onecomp.post_process.GlobalPTQ options: show_source: false

::: onecomp.post_process.GlobalPTQDistributed options: show_source: false

Block-wise PTQ

::: onecomp.post_process.BlockWisePTQ options: show_source: false

Router Fine-Tuning

::: onecomp.post_process.RouterFineTuning options: show_source: false

LoRA SFT

For the underlying adapter method, see LoRA. See the LoRA SFT guide for usage.

::: onecomp.post_process.PostProcessLoraSFT options: show_source: false

Convenience Variants

PostProcessLoraTeacherSFT and PostProcessLoraTeacherOnlySFT are pre-configured variants of PostProcessLoraSFT with different default loss weights:

::: onecomp.post_process.PostProcessLoraTeacherSFT options: show_source: false show_bases: false members: false

::: onecomp.post_process.PostProcessLoraTeacherOnlySFT options: show_source: false show_bases: false members: false