xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-16 12:05:57 +08:00
Code Issues Packages Projects Releases Wiki Activity

Avoid overwriting vllm_compile_cache.py (#17418)

Browse Source 
Signed-off-by: Keyun Tong <tongkeyun@gmail.com>
This commit is contained in:
Keyun Tong 2025-05-01 00:30:57 -07:00 committed by GitHub
parent 3c3d767201
commit 26bc4bbcd8
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 5 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
vllm/compilation/backends.py
View File

@ -45,6 +45,7 @@ class CompilerManager:
self.cache: Dict[Tuple[Optional[int], int, str], Any] = dict()
cls = InductorAdaptor if use_inductor else EagerAdaptor
self.compiler = cls()
self.is_cache_updated = False
def compute_hash(self, vllm_config: VllmConfig) -> str:
return self.compiler.compute_hash(vllm_config)
@ -66,11 +67,11 @@ class CompilerManager:
disable_cache=disable_cache)
def save_to_file(self):
if self.disable_cache:
if self.disable_cache or not self.is_cache_updated:
return
printer = pprint.PrettyPrinter(indent=4)
data = printer.pformat(self.cache)
with open(self.cache_file_path, "w") as f:
printer = pprint.PrettyPrinter(indent=4)
data = printer.pformat(self.cache)
f.write(data)
def load(self,
@ -131,6 +132,7 @@ class CompilerManager:
if handle is not None:
self.cache[(runtime_shape, graph_index,
self.compiler.name)] = handle
self.is_cache_updated = True
if graph_index == 0:
# adds some info logging for the first graph
logger.info("Cache the graph of shape %s for later use",