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 16, 05:45 UTC
Drop ROCm load format check (#10767)
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
wangxiyuan committed
Dec 4, 2024 at 12:32 UTC
b5b647b084de3a5a29d35ca527c9901f8e6a4e7e
1 file changed
+3
-20
vllm/config.py
+3
-20
@@ -931,7 +931,9 @@ class LoadConfig:
931
if isinstance(model_loader_extra_config, str):
932
self.model_loader_extra_config = json.loads(
933
model_loader_extra_config)
934
- self._verify_load_format()
934
+ if isinstance(self.load_format, str):
935
+ load_format = self.load_format.lower()
936
+ self.load_format = LoadFormat(load_format)
937
938
if self.ignore_patterns is not None and len(self.ignore_patterns) > 0:
939
logger.info(
@@ -940,25 +942,6 @@ class LoadConfig:
942
else:
943
self.ignore_patterns = ["original/**/*"]
944
943
- def _verify_load_format(self) -> None:
944
- if not isinstance(self.load_format, str):
945
- return
946
-
947
- load_format = self.load_format.lower()
948
- self.load_format = LoadFormat(load_format)
949
-
950
- rocm_not_supported_load_format: List[str] = []
951
- if current_platform.is_rocm(
952
- ) and load_format in rocm_not_supported_load_format:
953
- rocm_supported_load_format = [
954
- f for f in LoadFormat.__members__
955
- if (f not in rocm_not_supported_load_format)
956
- ]
957
- raise ValueError(
958
- f"load format '{load_format}' is not supported in ROCm. "
959
- f"Supported load formats are "
960
- f"{rocm_supported_load_format}")
961
-
945
946
@dataclass
947
class ParallelConfig: