save_checkpoint corrupted
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 9k
- Forks
- 1.5k
- Avg merge
- 2d 4h
- Merged PRs (30d)
- 3
Description
Saving a checkpoint when local_rank==0 will have one process per node writing to the same location and thus corrupt the checkpoint if nodes > 1
For example I got the error:
RuntimeError: storage has wrong size: expected 4249925191133211469 got 65536
Surely we would need to check if local_rank==0 and the node is the master-node, which the script currently doesn't do?
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Inspect examples/imagenet/main_amp.py around line 253 and trace how checkpoint saving is gated during multi-node distributed training. Reproduce the save with more than one node and verify that only the master node writes the checkpoint and that the resulting file loads without the reported storage-size error.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python, pytorch
- Domain
- distributed-systems, machine-learning
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100