xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-06-02 13:37:53 +08:00
Code Issues Packages Projects Releases Wiki Activity

mem

Browse Source
This commit is contained in:
Woosuk Kwon 2025-10-23 00:19:05 +00:00
parent a5281395e9
commit f65da69c72
1 changed files with 8 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
vllm/v1/worker/gpu_model_runner.py
View File

@ -509,6 +509,14 @@ class GPUModelRunner(LoRAModelRunnerMixin, KVConnectorModelRunnerMixin):
pin_memory=self.pin_memory, pin_memory=self.pin_memory,
) )
# device_id = self.device.index
# def cb(_device, _alloc, _device_alloc, _device_free):
# torch.cuda.memory._dump_snapshot(f"/tmp/vllm_oom_{device_id}.pickle")
# torch.cuda.memory._record_memory_history(max_entries=100_000)
# torch._C._cuda_attach_out_of_memory_observer(cb)
def reset_mm_cache(self) -> None: def reset_mm_cache(self) -> None:
if self.mm_budget: if self.mm_budget:
self.mm_budget.reset_cache() self.mm_budget.reset_cache()