[Optimization 5/n] Add Optimization Orchestrator #78
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
meta-cla
bot
added
the
CLA Signed
Jack-Khuu
reviewed
Jan 14, 2026
|
|
||
| if divergence > self.divergence_threshold: | ||
| self.logger.warning( | ||
| f"[{round_num}] ⚠️ EXCESSIVE DIVERGENCE: {new_time:.4f} ms is {divergence:.1f}% worse" |
Contributor
There was a problem hiding this comment.
nit: Excessive => Egregious?
Comment on lines
416
to
430
| if pytorch_baseline_time and pytorch_baseline_time != float("inf"): | ||
| pytorch_speedup = pytorch_baseline_time / best_time | ||
| self.logger.info(f" PyTorch baseline: {pytorch_baseline_time:.4f} ms") | ||
| self.logger.info( | ||
| f" Baseline time: {baseline_results['time_ms']:.4f} ms" | ||
| ) | ||
| self.logger.info(f" Best time: {best_time:.4f} ms") | ||
| self.logger.info(f" Speedup vs PyTorch: {pytorch_speedup:.2f}x") | ||
| self.logger.info(f" Speedup vs baseline: {baseline_speedup:.2f}x") | ||
| self.logger.info(f" Improvement: {improvement_percent:.1f}%") | ||
| else: | ||
| self.logger.info(f" Baseline time: {baseline_results['time_ms']:.4f} ms") | ||
| self.logger.info(f" Best time: {best_time:.4f} ms") | ||
| self.logger.info(f" Speedup: {baseline_speedup:.2f}x") | ||
| self.logger.info(f" Improvement: {improvement_percent:.1f}%") |
Contributor
There was a problem hiding this comment.
nit: Not necessarily this order, but just some light refactoring. It'd be nice if the conditional when there's a pytorch baseline (or any arbitrary additional baseline) just wraps it's unique lines
Suggested change
| if pytorch_baseline_time and pytorch_baseline_time != float("inf"): | |
| pytorch_speedup = pytorch_baseline_time / best_time | |
| self.logger.info(f" PyTorch baseline: {pytorch_baseline_time:.4f} ms") | |
| self.logger.info( | |
| f" Baseline time: {baseline_results['time_ms']:.4f} ms" | |
| ) | |
| self.logger.info(f" Best time: {best_time:.4f} ms") | |
| self.logger.info(f" Speedup vs PyTorch: {pytorch_speedup:.2f}x") | |
| self.logger.info(f" Speedup vs baseline: {baseline_speedup:.2f}x") | |
| self.logger.info(f" Improvement: {improvement_percent:.1f}%") | |
| else: | |
| self.logger.info(f" Baseline time: {baseline_results['time_ms']:.4f} ms") | |
| self.logger.info(f" Best time: {best_time:.4f} ms") | |
| self.logger.info(f" Speedup: {baseline_speedup:.2f}x") | |
| self.logger.info(f" Improvement: {improvement_percent:.1f}%") | |
| self.logger.info(f" Best time: {best_time:.4f} ms") | |
| self.logger.info(f" Baseline time: {baseline_results['time_ms']:.4f} ms") | |
| self.logger.info(f" Speedup vs baseline: {baseline_speedup:.2f}x") | |
| if pytorch_baseline_time and pytorch_baseline_time != float("inf"): | |
| pytorch_speedup = pytorch_baseline_time / best_time | |
| self.logger.info(f" PyTorch baseline: {pytorch_baseline_time:.4f} ms") | |
| self.logger.info(f" Speedup vs PyTorch: {pytorch_speedup:.2f}x") | |
| self.logger.info(f" Improvement: {improvement_percent:.1f}%") |
| "history": list(self.history), | ||
| } | ||
|
|
||
| def verify_with_refinement( |
Contributor
There was a problem hiding this comment.
I'm not ecstatic about having two functions (run/verify_with_refinement) that are this similar
Is there an easy way to pull out/tweak the single pass logic from run and reuse that?
added 27 commits
January 15, 2026 11:44
Consolidates previous kernel_benchmark.py and pytorch_benchmark.py into a streamlined 3-file architecture with clear separation of concerns: Architecture: - benchmark.py (299 lines): Main Benchmark class with simplified API - benchmark_kernel(): Always uses subprocess for crash protection - benchmark_pytorch(): Always uses direct mode for stable code - BenchmarkLockManager: GPU lock management for multi-worker scenarios - timing.py (437 lines): Complete timing infrastructure - Timing: time_with_cuda_events(), time_with_triton_do_bench() - Loading: prepare_pytorch_model(), load_kernel_function() - Stats: compute_timing_stats() with essential metrics (mean/std/min/max) - kernel_subprocess.py (442 lines): Subprocess runner for kernel isolation - Crash protection for potentially buggy kernels - Clean CUDA state between runs - Timeout handling Key improvements: - Eliminated string code generation (was generating Python as strings) - Removed unnecessary statistics (median, p25/p75/p95/p99) - Removed confusing use_subprocess parameter (behavior now deterministic) - Fixed dtype bug causing incorrect speedup measurements - Reduced from 5 files to 3 files with clearer naming - Code reduction: ~1,400 lines → 1,178 lines Simple API: bench = Benchmark(logger, temp_dir, lock, worker_id) pytorch_result = bench.benchmark_pytorch(problem_file) kernel_result = bench.benchmark_kernel(kernel_file, problem_file) speedup = pytorch_result['stats']['mean'] / kernel_result['time_ms']
kaiming-cheng
force-pushed
the
kaiming/opt_component_5
branch
from
be2accf to
dd55d1d
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary
Introduces
OptimizationOrchestratorthat iteratively optimizes Triton kernels using NCU profiling and bottleneck analysis.Add
verify_with_refinement()method toworker.pyfor single-pass verification with refinement loopHow
OptimizationOrchestratorWorksVerificationWorker)