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 15, 19:45 UTC
[Bugfix] Fix torch_xla in V0 which can't handle None seed introduced … (#14844)
Signed-off-by: Yarong Mu <ymu@google.com>
yarongmu-google committed
Mar 14, 2025 at 17:41 UTC
dd344e03425087ebcfc3f98f91821c7e5d316832
1 file changed
+3
vllm/worker/tpu_worker.py
+3
@@ -51,6 +51,9 @@ class TPUWorker(LoRANotSupportedWorkerBase, LocalOrDistributedWorkerBase):
51
self.model_runner: TPUModelRunner = TPUModelRunner(
52
vllm_config=vllm_config, is_driver_worker=is_driver_worker)
53
54
+ if self.model_config.seed is None:
55
+ self.model_config.seed = 0
56
+
57
def init_device(self) -> None:
58
os.environ["PJRT_DEVICE"] = "TPU"
59
torch.set_grad_enabled(False)