accelerate
7554afbc - Fix mutable default in Megatron init and IndexError on empty ModuleList (#3944)

Commit
17 days ago
Fix mutable default in Megatron init and IndexError on empty ModuleList (#3944) * Fix mutable default argument in Megatron init and IndexError on empty ModuleList In megatron_lm.py, initialize() used args_defaults={} as a default parameter, which is shared across calls and can cause unexpected state persistence. In other.py, is_repeated_block() accessed module[0] without checking if the ModuleList was empty, causing IndexError for empty ModuleLists. * Apply style fixes --------- Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Author
Parents
Loading