๐
๐
Old Age
Language Model Alignment with Elastic Reset
December 06, 2023 ยท Entered Twilight ยท ๐ Neural Information Processing Systems
Repo contents: .gitignore, LICENSE, README.md, stackllama
Authors
Michael Noukhovitch, Samuel Lavoie, Florian Strub, Aaron Courville
arXiv ID
2312.07551
Category
cs.CL: Computation & Language
Citations
39
Venue
Neural Information Processing Systems
Repository
https://github.com/mnoukhov/elastic-reset
โญ 5
Last Checked
2 months ago
Abstract
Finetuning language models with reinforcement learning (RL), e.g. from human feedback (HF), is a prominent method for alignment. But optimizing against a reward model can improve on reward while degrading performance in other areas, a phenomenon known as reward hacking, alignment tax, or language drift. First, we argue that commonly-used test metrics are insufficient and instead measure how different algorithms tradeoff between reward and drift. The standard method modified the reward with a Kullback-Lieber (KL) penalty between the online and initial model. We propose Elastic Reset, a new algorithm that achieves higher reward with less drift without explicitly modifying the training objective. We periodically reset the online model to an exponentially moving average (EMA) of itself, then reset the EMA model to the initial model. Through the use of an EMA, our model recovers quickly after resets and achieves higher reward with less drift in the same number of steps. We demonstrate that fine-tuning language models with Elastic Reset leads to state-of-the-art performance on a small scale pivot-translation benchmark, outperforms all baselines in a medium-scale RLHF-like IMDB mock sentiment task and leads to a more performant and more aligned technical QA chatbot with LLaMA-7B. Code available at github.com/mnoukhov/elastic-reset.
Community Contributions
Found the code? Know the venue? Think something is wrong? Let us know!
๐ Similar Papers
In the same crypt โ Computation & Language
๐
๐
Old Age
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
R.I.P.
๐ป
Ghosted
Language Models are Few-Shot Learners
R.I.P.
๐ป
Ghosted
RoBERTa: A Robustly Optimized BERT Pretraining Approach
R.I.P.
๐ป
Ghosted
BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension
R.I.P.
๐ป
Ghosted