github/tinygrad
1
1
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-04-29 03:00:14 -04:00
Code Issues Packages Projects Releases Wiki Activity

don't return loss that's scaled

Browse Source
This commit is contained in:
Francis Lata
2025-03-14 08:13:03 -07:00
parent f3fd2757a8
commit c4dc02b4ab
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
examples/mlperf/model_train.py
View File

@@ -395,9 +395,9 @@ def train_retinanet():
optim.zero_grad()
losses = model(normalize(x, GPUS), **kwargs)
loss = (sum([l for l in losses.values()]) * loss_scaler)
loss = sum([l for l in losses.values()])
loss.backward()
(loss * loss_scaler).backward()
for t in optim.params: t.grad = t.grad.contiguous() / loss_scaler
optim.step()