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

Expose PyTorch profiler configuration to environment variables (#21803)

Browse Source 
Signed-off-by: Csrayz <33659823+Csrayz@users.noreply.github.com>
This commit is contained in:
Csrayz
2025-07-30 10:46:31 +08:00
committed by GitHub
parent fb58e3a651
commit b917da442b
4 changed files with 60 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

29
vllm/envs.py
View File

@@ -80,6 +80,10 @@ if TYPE_CHECKING:
VLLM_PLUGINS: Optional[list[str]] = None
VLLM_LORA_RESOLVER_CACHE_DIR: Optional[str] = None
VLLM_TORCH_PROFILER_DIR: Optional[str] = None
VLLM_TORCH_PROFILER_RECORD_SHAPES: bool = False
VLLM_TORCH_PROFILER_WITH_PROFILE_MEMORY: bool = False
VLLM_TORCH_PROFILER_WITH_STACK: bool = True
VLLM_TORCH_PROFILER_WITH_FLOPS: bool = False
VLLM_USE_TRITON_AWQ: bool = False
VLLM_ALLOW_RUNTIME_LORA_UPDATING: bool = False
VLLM_SKIP_P2P_CHECK: bool = False
@@ -629,6 +633,31 @@ environment_variables: dict[str, Callable[[], Any]] = {
lambda: (None if os.getenv("VLLM_TORCH_PROFILER_DIR", None) is None else os
.path.expanduser(os.getenv("VLLM_TORCH_PROFILER_DIR", "."))),
# Enable torch profiler to record shapes if set
# VLLM_TORCH_PROFILER_RECORD_SHAPES=1. If not set, torch profiler will
# not record shapes.
"VLLM_TORCH_PROFILER_RECORD_SHAPES":
lambda: bool(os.getenv("VLLM_TORCH_PROFILER_RECORD_SHAPES", "0") != "0"),
# Enable torch profiler to profile memory if set
# VLLM_TORCH_PROFILER_WITH_PROFILE_MEMORY=1. If not set, torch profiler
# will not profile memory.
"VLLM_TORCH_PROFILER_WITH_PROFILE_MEMORY":
lambda: bool(
os.getenv("VLLM_TORCH_PROFILER_WITH_PROFILE_MEMORY", "0") != "0"),
# Enable torch profiler to profile stack if set
# VLLM_TORCH_PROFILER_WITH_STACK=1. If not set, torch profiler WILL
# profile stack by default.
"VLLM_TORCH_PROFILER_WITH_STACK":
lambda: bool(os.getenv("VLLM_TORCH_PROFILER_WITH_STACK", "1") != "0"),
# Enable torch profiler to profile flops if set
# VLLM_TORCH_PROFILER_WITH_FLOPS=1. If not set, torch profiler will
# not profile flops.
"VLLM_TORCH_PROFILER_WITH_FLOPS":
lambda: bool(os.getenv("VLLM_TORCH_PROFILER_WITH_FLOPS", "0") != "0"),
# If set, vLLM will use Triton implementations of AWQ.
"VLLM_USE_TRITON_AWQ":
lambda: bool(int(os.getenv("VLLM_USE_TRITON_AWQ", "0"))),