ROCm Docker build fails after GPTQ support

[WoosukKwon]

hey, i just noticed the recent most push fails on rocm

vllm/setup.py

Line 222 in b81a6a6

"csrc/quantization/gptq/q_gemm.cu",

can you add this line in the setup.py along with awq quantization like so

vllm/setup.py

Line 228 in b81a6a6

vllm_extension_sources.append("csrc/quantization/awq/gemm_kernels.cu")

If this feature is suppose to support rocm can you also check as hipify is not able to import #include<hipblas.h>

Thank you

Originally posted by @hex-plex in #916 (comment)