ROCm `F.layer_norm` silently produces wrong output for large inputs (even below `M*N > 2^32` threshold)

May 21, 2026 · #184826
View on GitHub
Python Difficulty: Medium

Labels

triage review module: rocm has workaround module: correctness (silent) module: norms and normalization
Ready to work on this? Walk through the full fork-to-PR workflow so your first contribution goes smoothly.

Sign in required

Authenticate to use favourites & bookmarks

5