dnrops
/
transformers
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fix duplicate call to save_checkpoint when using deepspeed (#14946) 

* Fix duplicate call to save_checkpoint when using deepspeed / stage3_gather_fp16_weights_on_model_save

* Revert "Fix duplicate call to save_checkpoint when using deepspeed / stage3_gather_fp16_weights_on_model_save"

This reverts commit 6a3dec0397.

* Delete correct duplicate invocation of deepspeed save_checkpoint
This commit is contained in:
Mihai Balint 2021-12-27 21:25:26 +02:00 committed by GitHub
parent 03885a3f50
commit c1138273d4
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 0 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
src/transformers/trainer.py
View File

@ -1999,9 +1999,6 @@ class Trainer:
# This must be called on all ranks
self.deepspeed.save_fp16_model(output_dir, WEIGHTS_NAME)
# save a deepspeed checkpoint as well (this is very fast)
self.deepspeed.save_checkpoint(output_dir)
elif self.args.should_save:
self._save(output_dir)