[plugin] oot torch profiler activity support

[DevashishLal-CB]

Motivation

Currently integrating HW specific torch profile activities required in tree changes, which can easily be driven by the device mixin from the platform plugins system

Modifications

extends device mixin and and profiler utilities
to accept oot torch profiler activities

Accuracy Tests

Speed Tests and Profiling

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

Review and Merge Process

1. Ping Merge Oncalls to start the process. See the PR Merge Process.
2. Get approvals from CODEOWNERS and other reviewers.
3. Trigger CI tests with comments or contact authorized users to do so.
   • Common commands include /tag-and-rerun-ci, /tag-run-ci-label, /rerun-failed-ci
4. After green CI and required approvals, ask Merge Oncalls or people with Write permission to merge the PR.

---

CI States

Latest PR Test (Base): ❌ Run #29622622859
Latest PR Test (Extra): ❌ Run #29622622748

[DevashishLal-CB]

/tag-and-rerun-ci

[gemini-code-assist]

Code Review

This pull request introduces support for out-of-tree (OOT) platforms in the profiler by adding get_torch_profiler_activity_str and get_torch_profiler_activity methods to DeviceMixin, and updating the profiler manager and utilities to conditionally load these custom activities. The review feedback suggests returning None instead of raising NotImplementedError by default to avoid forcing all OOT platforms to implement these methods. Additionally, the feedback points out several potential runtime issues (such as TypeError or dictionary key errors) if these methods return None, and provides code suggestions to safely handle optional values.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

Instead of raising NotImplementedError by default, it is safer to return None (using Optional). This prevents out-of-tree (OOT) platforms that do not require custom profiler activities from being forced to implement these methods, avoiding potential runtime crashes.

Suggested change

	def get_torch_profiler_activity_str(self) -> str:
	"""[Planned] Return the torch profiler activity string."""
	raise NotImplementedError
	def get_torch_profiler_activity(self) -> torch.profiler.ProfilerActivity:
	"""[Planned] Return the torch profiler activity."""
	raise NotImplementedError
	def get_torch_profiler_activity_str(self) -> Optional[str]:
	"""[Planned] Return the torch profiler activity string."""
	return None
	def get_torch_profiler_activity(self) -> Optional[torch.profiler.ProfilerActivity]:
	"""[Planned] Return the torch profiler activity."""
	return None

[gemini-code-assist]

If get_torch_profiler_activity_str() returns None (or if it is not implemented), calling hasattr(torch.profiler.ProfilerActivity, None) will raise a TypeError. We should retrieve the values first and check if they are not None before checking attributes and updating the map.

Suggested change

	if current_platform.is_out_of_tree():
	if hasattr(
	torch.profiler.ProfilerActivity,
	current_platform.get_torch_profiler_activity_str(),
	):
	activity_map[current_platform.get_torch_profiler_activity_str()] = (
	current_platform.get_torch_profiler_activity()
	)
	if current_platform.is_out_of_tree():
	activity_str = current_platform.get_torch_profiler_activity_str()
	activity = current_platform.get_torch_profiler_activity()
	if activity_str and activity and hasattr(torch.profiler.ProfilerActivity, activity_str):
	activity_map[activity_str] = activity

[gemini-code-assist]

To prevent potential TypeError or issues when get_torch_profiler_activity_str() returns None, we should check that the returned activity string is not None before checking if it is in activities.

Suggested change

	if current_platform.is_out_of_tree():
	if current_platform.get_torch_profiler_activity_str() in activities:
	inners.append(
	_ProfilerTorch(
	**kwargs,
	activities=activities,
	with_stack=with_stack,
	record_shapes=record_shapes,
	)
	)
	if current_platform.is_out_of_tree():
	activity_str = current_platform.get_torch_profiler_activity_str()
	if activity_str and activity_str in activities:
	inners.append(
	_ProfilerTorch(
	**kwargs,
	activities=activities,
	with_stack=with_stack,
	record_shapes=record_shapes,
	)
	)

[gemini-code-assist]

If get_torch_profiler_activity_str() returns None, this will set activity_map[None] = None, which can cause issues later. We should check that both the activity string and the activity itself are not None before adding them to the map.

Suggested change

	if current_platform.is_out_of_tree():
	activity_map[current_platform.get_torch_profiler_activity_str()] = (
	current_platform.get_torch_profiler_activity()
	)
	if current_platform.is_out_of_tree():
	activity_str = current_platform.get_torch_profiler_activity_str()
	activity = current_platform.get_torch_profiler_activity()
	if activity_str and activity:
	activity_map[activity_str] = activity

[DevashishLal-CB]

/tag-and-rerun-ci

[DevashishLal-CB]

/rerun-failed-ci