Fix Nebula checkpoint engine commit() API mismatch (#7740)
## Summary
- Fix `AttributeError: 'str' object has no attribute 'tag'` when using
Nebula checkpoint engine
- Pass `CheckpointCommitInfo` object instead of raw `tag` string to
`checkpoint_engine.commit()`
## Description
The `CheckpointEngine.commit()` interface expects a
`CheckpointCommitInfo` object, but two call sites in `engine.py` were
passing a raw `tag` string instead:
1. `save_checkpoint()` at line 3695
2. `save_16bit_model()` at line 4230
This worked with `TorchCheckpointEngine` because it ignores the
parameter, but `NebulaCheckpointEngine` accesses `info.tag`, causing the
crash.
## Changes
- **Line 3695**: Create `CheckpointCommitInfo` object before calling
`commit()`
- **Line 4231**: Use existing `commit_info` variable instead of `tag`
## Test plan
- [ ] Verify Nebula checkpoint engine saves work without
`AttributeError`
- [ ] Verify TorchCheckpointEngine still works (no regression)
- [ ] Run existing checkpoint-related unit tests
Fixes #7678
Signed-off-by: Rakshit-gen <sisodiarakshit456@gmail.com>
Co-authored-by: Olatunji Ruwase <tunji.ruwase@snowflake.com>
