Skip to content

[torch.compile] Use custom ops when use_inductor=False #19618

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Jun 13, 2025
Merged

[torch.compile] Use custom ops when use_inductor=False #19618

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
18 changes: 15 additions & 3 deletions vllm/config.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4450,15 +4450,27 @@ def __post_init__(self):
self.compilation_config.custom_ops.append("+rms_norm")
if envs.VLLM_USE_V1 and self.model_config is not None and \
not self.model_config.enforce_eager:
# FIXME(rob): Add function to set all of these.
if not self.compilation_config.custom_ops:
self.compilation_config.custom_ops = ["none"]
# By default, V1 uses piecewise CUDA graphs. If full_cuda_graph
# is set to True, full CUDA graphs will be used.
self.compilation_config.cudagraph_num_of_warmups = 1
self.compilation_config.pass_config.enable_fusion = False
self.compilation_config.pass_config.enable_noop = False
self.compilation_config.level = CompilationLevel.PIECEWISE
self.compilation_config.set_splitting_ops_for_v1()

# The behavior of custom ops with inductor depends on the config:
# - If use_inductor=True and custom_ops is empty:
# Inductor generates Triton kernels for all registered custom ops
# (default behavior)
# - If use_inductor=True and custom_ops is non-empty:
# Custom CUDA kernels are used for specified ops while inductor
# generates Triton kernels for remaining ops, including misc torch
# ops in the model.
Comment on lines +4461 to +4468
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

FYI @BoyuanFeng this might change how we do benchmarking

if (not self.compilation_config.custom_ops
and self.compilation_config.use_inductor):
# Let inductor generate Triton kernels for the custom ops.
self.compilation_config.custom_ops = ["none"]

self._set_cudagraph_sizes()

if self.cache_config.cpu_offload_gb > 0 and \
Expand Down