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
245 stars 20 forks source link

[WIP] hadamard support #222

Open wenhuach21 opened 2 months ago

wenhuach21 commented 2 months ago

negative impact with AutoRound in W2 and W4A4 MXFP4 for llama3