AI Dynamics

Global AI News Aggregator

About

Unsloth Monkey-Patching Causes Distributed Training Deadlock Issues

Eventually tracked down the problem to the Unsloth monkey-patching. Distributed runs even with one process fail, but if you use the regular models from transformers and `pef`, trained with `trl` then it doesn't deadlock… Hmm.

→ View original post on X — @alexjc