Han Zhu ab91112909

Improve infinity-check (#1862) ...

1. Attach the inf-check hooks if the grad scale is getting too small.
2. Add try-catch to avoid OOM in the inf-check hooks.
3. Set warmup_start=0.1 to reduce chances of divergence

2025-01-09 15:05:38 +08:00

..

ASR

Improve infinity-check (#1862)

2025-01-09 15:05:38 +08:00

SSL

Revert "Replace deprecated pytorch methods (#1814)" (#1841)

2024-12-18 16:49:57 +08:00

WSASR

Revert "Replace deprecated pytorch methods (#1814)" (#1841)

2024-12-18 16:49:57 +08:00