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 14, 05:30 UTC
[Misc] Remove pass_config from CompilationConfig dump_json excluded (#21911)
Signed-off-by: elvischenv <219235043+elvischenv@users.noreply.github.com>
elvischenv committed
Aug 5, 2025 at 10:17 UTC
7b455cf1c036d12470374d716800d0fd09290a5a
1 file changed
+9
-1
vllm/config.py
+9
-1
@@ -4374,12 +4374,20 @@ class CompilationConfig:
4374
"disabled_custom_ops": True,
4375
"compilation_time": True,
4376
"bs_to_padded_graph_size": True,
4377
- "pass_config": True,
4377
"traced_files": True,
4378
"inductor_compile_config": {
4379
"post_grad_custom_post_pass": True,
4380
},
4381
}
4382
+
4383
+ # exclude default attr in pass_config
4384
+ pass_config_exclude = {}
4385
+ for attr, default_val in vars(PassConfig()).items():
4386
+ if getattr(self.pass_config, attr) == default_val:
4387
+ pass_config_exclude[attr] = True
4388
+ if pass_config_exclude:
4389
+ exclude["pass_config"] = pass_config_exclude
4390
+
4391
# The cast to string is necessary because Pydantic is mocked in docs
4392
# builds and sphinx-argparse doesn't know the return type of decode()
4393
return str(