Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Improved Evaluations #2

Merged
merged 7 commits into from
Jun 1, 2024
Merged

Improved Evaluations #2

merged 7 commits into from
Jun 1, 2024

Conversation

farzadab
Copy link
Contributor

@farzadab farzadab commented May 31, 2024

This PR does the following:

  • Bugfix: AnyInstruct was not using val/train splits (as it had somehow gotten removed from the last PR)
  • Faster evaluations by using DDP multiprocessing for text generation
  • Validation loss added by default to training

@farzadab farzadab changed the title Evaluations Improved Evaluations May 31, 2024
@farzadab farzadab requested a review from juberti May 31, 2024 21:13
ultravox/training/train.py Outdated Show resolved Hide resolved
ultravox/data/datasets.py Show resolved Hide resolved
ultravox/training/config_base.py Show resolved Hide resolved
ultravox/training/train.py Outdated Show resolved Hide resolved
@farzadab
Copy link
Contributor Author

Small training run to check for regressions seems to have run fine: https://wandb.ai/fixie/ultravox/runs/q3rh3yxv

@farzadab farzadab merged commit 87eb25d into main Jun 1, 2024
1 check passed
@farzadab farzadab deleted the farzad-evals branch June 1, 2024 00:20
@farzadab farzadab restored the farzad-evals branch June 3, 2024 16:40
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants