Files
rustytorch/crates/core/rtx-autograd/src
Omar SobhandClaude Fable 5 38645c7c74 fix(autograd): AbsBackward produced NaN for exactly-zero inputs
sign was computed as x/|x|, which is 0/0 = NaN at x = 0; one zero element
in an |pred - target| loss poisoned every upstream gradient (hit
deterministically by dg-gnn HetGAT training). Compute x/(|x| + tiny)
instead so sign(0) = 0 (the subgradient convention). Regression test
included.

Co-Authored-By: Claude Fable 5 <[email protected]>
2026-08-20 11:53:30 -07:00
..
2026-03-04 00:08:42 +00:00
2026-03-04 00:08:42 +00:00
2026-03-04 00:08:42 +00:00
2026-03-04 00:08:42 +00:00
2026-03-04 00:08:42 +00:00
2026-03-04 00:08:42 +00:00
2026-03-04 00:08:42 +00:00