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 4, 12:30 UTC
[Bugfix] Report compile time for in-memory cache hit path (#41023)
Signed-off-by: Frederik Gossen <frgossen@meta.com>
Frederik Gossen committed
Apr 29, 2026 at 11:32 UTC
a05848e255614401e3813c656b8cfa94969952d4
1 file changed
+2
-1
vllm/compilation/backends.py
+2
-1
@@ -279,6 +279,7 @@ class CompilerManager:
279
compilation_counter.num_backend_compilations += 1
280
281
compiled_graph = None
282
+ handle = None
283
284
# try to load from the cache
285
compiled_graph = self.load(graph, example_inputs, graph_index, compile_range)
@@ -357,7 +358,7 @@ class CompilerManager:
358
)
359
except StopCompiling:
360
assert cache_key is not None
360
- return self.loaded_artifacts[cache_key]
361
+ compiled_graph = self.loaded_artifacts[cache_key]
362
if cache_key is not None and compiled_graph is not None:
363
self.loaded_artifacts[cache_key] = compiled_graph
364