Skip to content
/ GAP Public

[ACL 2023] Gradient Ascent Post-training Enhances Language Model Generalization

License

Notifications You must be signed in to change notification settings

kaistAI/GAP

Repository files navigation

GAP

[ACL 2023] Gradient Ascent Post-training Enhances Language Model Generalization

Dongkeun Yoon*, Joel Jang*, Sungdong Kim, Minjoon Seo (* Equal Contribution)

Updating language models towards divergence can lead to better performance.

1. Run GAP

The below example command will run 100 GAP runs with Github (Pile) on OPT-350M. You can modify the script or the config to change the training data, model etc.

bash scripts/run_git.sh

2. Test some of our top performing GAP runs

The below runs resulted in substantial performance gains in dialogue generation tasks. Note that performance gains in dialogue tasks doesn't necessarily correlate with performance gains in classification tasks, and the following models may result in performance drops in classification tasks.

  • OPT-350M
    python run.py --config configs/final/extraction-350.json --index 79 --num_train_epochs 8 --check_val_every_n_epoch 8
  • OPT-1.3B
    python run.py --config configs/final/github-1.3.json --index 56 --num_train_epochs 7 --check_val_every_n_epoch 7
  • OPT-2.7B
    python run.py --config configs/final/cc-2.7.json --index 39 --num_train_epochs 7 --check_val_every_n_epoch 7

3. Check the qualitative results

The generation results for some GAP models are available at qualitative.csv.

target: The gold response.

opt-1.3b: Baseline.

opt-1.3b-github_40, opt-1.3b-cc_61: GAP models.

4. Conduct your own analysis

We provide the full evaluation results in full_results/. These results can be aligned with the training dataset files in data/train_data/ using the corpora and index column. For example if corpora == Git. and index == 0, the evaluation result comes from a model trained on the 0th entry of data/train_data/github.csv.

About

[ACL 2023] Gradient Ascent Post-training Enhances Language Model Generalization

Resources

License

Stars

Watchers

Forks