trunk/83441a98cd065f2a334a0b8c70bf0c00f0e4cd2e: Set gradcheck_nondet_tol for interpolate linear OpInfo (#199901)
- PyTorch: 2083 events in the last 90 days
- PyTorch: 2054th Release in the last 90 days
- Previous: earlier the same day · trunk/37f6057e0908a3b2ab972cf0b17a1bcfc1ba2564: [typing] Replace Any with object in auto-chunker metadata (#200333)
What happened
Go through the following checklist Passes lint ( spin fixlint ) Added/updated tests Updated documentation (if applicable) Included benchmark results (for PRs impacting perf) Linked issue or supporting maintainer Fixes #197512 BC-breaking? No Summary (human written only) CUDA backward uses atomic additions, leading to non-deterministic accumulation order and ~1 ULP drift in float64 across runs. Gradients are verified…
Summary assembled by rule from the sources below