Lightning-AI / Lightning-AI/pytorch-lightning
Wandb 1.x step handling
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 31.4k
- Forks
- 3.8k
- Avg merge
- 6d 7h
- Merged PRs (30d)
- 6
Description
### Description & Motivation
I recently just upgraded from lightning 1.x to 2.x. During the change, it seems that lightning has changed the way that steps are handled when logging to wandb. In particular, instead of using wandb's `step` argument, as was done in 1.x, lightning 2.x's step is logged as `trainer/global_step`. The trouble for me is that this makes comparing runs between 1.x and 2.x very difficult, because the steps are not on the same scale.
### Pitch
I would like an option to `WandbLogger` that allows me to optionally configure the 1.x step behavior.
### Alternatives
_No response_
### Additional context


cc @lantiga @borda @morganmcg1 @borisdayma @scottire @parambharat
Contributor guide
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
Start by locating WandbLogger and examining how logging steps are passed to Weights & Biases in the current version, then compare that behavior with the 1.x behavior described in the issue. Done means an optional configuration restores compatible 1.x step handling without changing the default behavior, with coverage for both modes.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- machine-learning
- Issue type
- Feature
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 38/100