admin/lerobot
1
0
Fork 0
mirror of https://github.com/huggingface/lerobot.git synced 2026-05-17 01:30:14 +00:00
Code Issues Packages Projects Releases Wiki Activity

refactor(profiling): shrink lerobot_train.py diff via start()/finalize()

Browse Source 
Replace the `with profiler or nullcontext():` wrap around the entire
training loop with explicit `profiler.start()` / `profiler.finalize()`
calls, and tighten `_section(...)` regions in `update_policy` to only
wrap the hot calls (forward / backward / optimizer.step).

This avoids ~120 lines of pure re-indentation noise while keeping the
exact same artifacts on disk and the same public behavior.

lerobot_train.py diff vs main: 267 -> 29 changed lines.

Made-with: Cursor
This commit is contained in:
Pepijn
2026-04-17 10:59:43 +01:00
parent a6dd28e8b4
commit 1ac8e96575
3 changed files with 128 additions and 131 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/scripts/test_model_profiling.py
+6 -6
View File
@@ -400,12 +400,12 @@ def test_training_profiler_section_records_duration(tmp_path):
output_dir=tmp_path,
device=torch.device("cpu"),
)
with profiler:
with profiler.section("forward"):
pass
with profiler.section("backward"):
pass
profiler.step(1, argparse.Namespace(update_s=0.5, dataloading_s=0.01))
profiler.start()
with profiler.section("forward"):
pass
with profiler.section("backward"):
pass
profiler.step(1, argparse.Namespace(update_s=0.5, dataloading_s=0.01))
profiler.finalize()
payload = json.loads((tmp_path / "step_timing_summary.json").read_text())