trunk/23047d0839b1b5ef8254539f46f191c44659af97: Fix oversized storage for reduced Smooth L1 loss (#200894)
- PyTorch: 2313 events in the last 90 days
- PyTorch: 2283th Release in the last 90 days
- Previous: earlier the same day · trunk/296101d35a7ab06f3cc8a4c3ec70f4bb020d5a9e: [caffe2] Use thread-local cache for N-bit embedding bags (#200898)
What happened
Linked issue or supporting maintainer Part of #185647 BC-breaking? No. Summary (human written only) F.smooth_l1_loss has the same oversized scalar storage issue fixed for mse_loss in #199823 . This PR uses the same reduced output pattern to Smooth L1, so mean and sum reductions no longer retain the full elementwise buffer. Agent details (if applicable) I used Codex to help implement this fix. I reviewed the diff and…
Summary assembled by rule from the sources below