A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 17, 05:30 UTC
[ROCm] Temporarily remove GPTQ ROCm support (#2138)
Woosuk Kwon committed
Dec 15, 2023 at 17:13 UTC
2acd76f346efcdff4f6ca1d92fe1575c448e4b70
2 files changed
+2
-2
setup.py
+1
-1
@@ -219,13 +219,13 @@ vllm_extension_sources = [
219
"csrc/activation_kernels.cu",
220
"csrc/layernorm_kernels.cu",
221
"csrc/quantization/squeezellm/quant_cuda_kernel.cu",
222
- "csrc/quantization/gptq/q_gemm.cu",
222
"csrc/cuda_utils_kernels.cu",
223
"csrc/pybind.cpp",
224
]
225
226
if _is_cuda():
227
vllm_extension_sources.append("csrc/quantization/awq/gemm_kernels.cu")
228
+ vllm_extension_sources.append("csrc/quantization/gptq/q_gemm.cu")
229
230
vllm_extension = CUDAExtension(
231
name="vllm._C",
vllm/config.py
+1
-1
@@ -143,7 +143,7 @@ class ModelConfig:
143
144
def _verify_quantization(self) -> None:
145
supported_quantization = ["awq", "gptq", "squeezellm"]
146
- rocm_not_supported_quantization = ["awq"]
146
+ rocm_not_supported_quantization = ["awq", "gptq"]
147
if self.quantization is not None:
148
self.quantization = self.quantization.lower()
149