tocmo0nlord/axolotl
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

fix checkpints on multigpu (#481)

Browse Source
This commit is contained in:
Wing Lian
2023-08-26 12:00:03 -04:00
committed by GitHub
parent 56c4a94caf
commit 31f3e71764
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/axolotl/monkeypatch/relora.py
View File

@@ -131,7 +131,7 @@ class ReLoRACallback(TrainerCallback):
and state.global_step % self.relora_steps != 0 and state.global_step % self.relora_steps != 0
): ):
if self.quantized: if self.quantized:
if self.last_full_model != checkpoint_folder: if is_main_process() and self.last_full_model != checkpoint_folder:
# ensure the latest full parameter save is in the latest checkpoint # ensure the latest full parameter save is in the latest checkpoint
# folder, so that automatic pruning of checkpoints does not remove it # folder, so that automatic pruning of checkpoints does not remove it
LOG.info(f"moving last full parameter save to {checkpoint_folder}") LOG.info(f"moving last full parameter save to {checkpoint_folder}")