iHore/vLLM
8
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Revert compile-cache device index regression on CPU (#53304)

Browse Source
This commit is contained in:
Kevin H. Luu
2026-08-21 16:08:13 -04:00
committed by GitHub
parent f15ea66b30
commit c0ff33404b
2 changed files with 2 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files
vllm/compilation/backends.py
+1 -4
View File
@@ -1074,10 +1074,7 @@ class VllmBackend:
self.compilation_config.cache_dir = cache_dir
rank = vllm_config.parallel_config.rank
dp_rank = vllm_config.parallel_config.data_parallel_index
dev = torch.accelerator.current_device_index()
local_cache_dir = os.path.join(
cache_dir, f"rank_{rank}_{dp_rank}_dev{dev}", self.prefix
)
local_cache_dir = os.path.join(cache_dir, f"rank_{rank}_{dp_rank}", self.prefix)
os.makedirs(local_cache_dir, exist_ok=True)
self.compilation_config.local_cache_dir = local_cache_dir
vllm/compilation/decorators.py
+1 -2
View File
@@ -559,8 +559,7 @@ def _support_torch_compile(
rank = self.vllm_config.parallel_config.rank
dp_rank = self.vllm_config.parallel_config.data_parallel_index
dev = torch.accelerator.current_device_index()
cache_dir = os.path.join(cache_dir, f"rank_{rank}_{dp_rank}_dev{dev}")
cache_dir = os.path.join(cache_dir, f"rank_{rank}_{dp_rank}")
aot_compilation_path = os.path.join(cache_dir, "model")
if not envs.VLLM_DISABLE_COMPILE_CACHE:
loaded_fn = _try_load_aot_compiled_fn(self, aot_compilation_path)