search

HandH1998 / QQQ

QQQ is an innovative and hardware-optimized W4A8 quantization solution for LLMs.
https://arxiv.org/pdf/2406.09904
91 stars 8 forks source link

bugs: qqq_gemm.cu(183): error: identifier "__hfma2" is undefined #18

Closed Andy0422 closed 1 month ago

Andy0422 commented 2 months ago

Hello, could you give me a hand on this bugs? It seems the cuda version problem... My nvcc-V is 12.3, and GPU is H100, torch version is 2.3, python3.10 could you let me know your platform setting

HandH1998 commented 2 months ago

Hello,

could you give me a hand on this bugs? It seems the cuda version problem...

My nvcc-V is 12.3, and GPU is H100, torch version is 2.3, python3.10

could you let me know your platform setting

please refer to README