biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

workaround of AWQ for Turing GPUs (#1252)

Browse Source
This commit is contained in:
twaka
2023-10-11 11:48:16 +09:00
committed by GitHub
parent 91fce82c6f
commit 8285736840
3 changed files with 73 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
csrc/quantization/awq/dequantize.cuh
View File

@@ -16,7 +16,7 @@ namespace awq {
__device__ uint4 dequantize_s4_to_fp16x2(uint32_t const& source)
{
#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ < 800
#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ < 750
assert(false);
#else
uint4 result;