vllm_omni.profiler ¶

Modules:

Name	Description
`omni_torch_profiler`

OmniTorchProfilerWrapper ¶

Bases: WorkerProfiler

Base torch profiler wrapper with platform-agnostic functionality.

Provides common profiler features: - Custom trace file naming with stage/rank info - Background gzip compression via subprocess - Returns trace file paths from get_results() for orchestrator collection

Subclasses can override hook methods for platform-specific behavior: - _get_default_activities(): Return default activities for the platform - _create_profiler(): Create platform-specific profiler instance - _on_stop_hook(): Handle platform-specific post-stop logic

dump_cpu_time_total `instance-attribute` ¶

dump_cpu_time_total = (
    "CPU" in activities and len(activities) == 1
)

local_rank `instance-attribute` ¶

local_rank = local_rank

profiler `instance-attribute` ¶

profiler = _create_profiler(profiler_config, activities)

profiler_config `instance-attribute` ¶

profiler_config = profiler_config

annotate_context_manager ¶

annotate_context_manager(name: str)

get_results ¶

get_results() -> dict

Return collected trace and table paths after stop.

set_trace_filename ¶

set_trace_filename(filename: str) -> None

Set the trace filename before starting profiling.

Parameters:

Name	Type	Description	Default
`filename`	`str`	Base filename without extension or rank suffix. e.g. "stage_0_llm_1234567890" Can also be a full path (e.g. from diffusion engine).	required

create_omni_profiler ¶

create_omni_profiler(
    profiler_config: ProfilerConfig,
    worker_name: str,
    local_rank: int,
    activities: list[TorchProfilerActivity] | None = None,
) -> OmniTorchProfilerWrapper

Factory function to create platform-specific profiler.

Uses the current platform's get_profiler_cls() to determine which profiler class to instantiate.

Parameters:

Name	Type	Description	Default
`profiler_config`	`ProfilerConfig`	Profiler configuration.	required
`worker_name`	`str`	Name of the worker.	required
`local_rank`	`int`	Local rank of the worker.	required
`activities`	`list[TorchProfilerActivity] \| None`	Optional list of profiler activities.	`None`

Returns:

Type	Description
`OmniTorchProfilerWrapper`	Platform-specific profiler instance.