Skip to content
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.

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

Improve warm-up and timing printing for eval #415

Merged
merged 1 commit into from
Dec 13, 2024
Merged
Changes from all commits
Commits
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: 11 additions & 7 deletions src/metatrain/cli/eval.py
Original file line number Diff line number Diff line change
Expand Up @@ -212,11 +212,15 @@ def _eval_targets(
total_time = 0.0
timings_per_atom = []

# Warm up with a single batch 5 times (to get accurate timings later)
batch = next(iter(dataloader))
systems = batch[0]
systems = [system.to(dtype=dtype, device=device) for system in systems]
for _ in range(5):
# Warm up with a single batch 10 times (to get accurate timings later).
# We use different batches to warm up torch potentially with different
# tensor sizes, so dynamic shape compilation happens. We have to cycle
# the dataloader in case there are few batches.
cycled_dataloader = itertools.cycle(dataloader)
for _ in range(10):
batch = next(cycled_dataloader)
systems = batch[0]
systems = [system.to(dtype=dtype, device=device) for system in systems]
evaluate_model(
model,
systems,
Expand Down Expand Up @@ -282,8 +286,8 @@ def _eval_targets(
std_per_atom = np.std(timings_per_atom)
logger.info(
f"evaluation time: {total_time:.2f} s "
f"[{1000.0*mean_per_atom:.2f} ± "
f"{1000.0*std_per_atom:.2f} ms per atom]"
f"[{1000.0*mean_per_atom:.4f} ± "
f"{1000.0*std_per_atom:.4f} ms per atom]"
)

if return_predictions:
Expand Down