Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Add runtime error in train.py if yaml config is improperly formatted …
…with extraneous or missing values (#506) ## Description This PR enables us to sanity check our train yaml configuration files before we run the full training pipeline. This enables us to catch errors in the YAML config before a training run starts. If a yaml config is improperly formatted with extraneous or missing values, a `NameError` will be thrown. ## Unit Test: Added unit tests to make sure we raise a `NameError` and/or warn the user if the yaml is incorrectly formatted. Warnings are used if the parameter has an optional default value. ## Integration Test: Before and after training runs show the same loss curves throughout training. <img width="333" alt="Screenshot 2023-08-15 at 10 57 55 AM" src="https://github.com/mosaicml/llm-foundry/assets/13524881/a0329a77-c54b-4d0b-961b-058e35021664"> **mpt-125m training branch master**: https://wandb.ai/mosaic-ml/chuck-runs/runs/2jq1uy86 **mpt-125m training branch chuck/add_yaml_sanity_check_train**: https://wandb.ai/mosaic-ml/chuck-runs/runs/n8wlqwec All runs at: https://wandb.ai/mosaic-ml/chuck-runs?workspace=user-chuck-tang98 --------- Co-authored-by: Daniel King <[email protected]>
- Loading branch information