large discrepancy between GPTQ model and qdq model

intel / auto-round

Advanced Quantization Algorithm for LLMs. This is official implementation of "Optimize Weight Rounding via Signed Gradient Descent for the Quantization of LLMs"

https://arxiv.org/abs/2309.05516

Apache License 2.0

172 stars 20 forks source link

large discrepancy between GPTQ model and qdq model #108

Closed wenhuach21 closed 2 months ago

wenhuach21 commented 3 months ago

waitting for the fix https://github.com/AutoGPTQ/AutoGPTQ/pull/640

wenhuach21 commented 2 months ago

fix in autoroun format