biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Perf] Cache xpu_get_mem_info() result to avoid duplicate calls (#32983)

Browse Source 
Signed-off-by: 7. Sun <jhao.sun@gmail.com>
This commit is contained in:
7. Sun
2026-01-24 04:56:23 +00:00
committed by GitHub
parent d0cbac5827
commit 14d03b8ddb
1 changed files with 2 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
vllm/v1/worker/xpu_worker.py
View File

@@ -107,7 +107,8 @@ class XPUWorker(Worker):
torch.xpu.empty_cache() torch.xpu.empty_cache()
torch_allocated_bytes = torch.xpu.memory_stats()["allocated_bytes.all.current"] torch_allocated_bytes = torch.xpu.memory_stats()["allocated_bytes.all.current"]
total_allocated_bytes = self.xpu_get_mem_info()[1] - self.xpu_get_mem_info()[0] free_mem, total_mem = self.xpu_get_mem_info()
total_allocated_bytes = total_mem - free_mem
non_torch_allocations = total_allocated_bytes - torch_allocated_bytes non_torch_allocations = total_allocated_bytes - torch_allocated_bytes
if non_torch_allocations > 0: if non_torch_allocations > 0: