search

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
247 stars 20 forks source link

fix gradient accumulate issue for varying input length #305

Closed wenhuach21 closed 6 days ago

wenhuach21 commented 1 week ago

https://unsloth.ai/blog/gradient