Scalable Ensembling For Mitigating Reward Overoptimisation
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866916293795381248 |
|---|---|
| author | Ahmed, Ahmed M. Rafailov, Rafael Sharkov, Stepan Li, Xuechen Koyejo, Sanmi |
| author_facet | Ahmed, Ahmed M. Rafailov, Rafael Sharkov, Stepan Li, Xuechen Koyejo, Sanmi |
| contents | Reinforcement Learning from Human Feedback (RLHF) has enabled significant advancements within language modeling for powerful, instruction-following models. However, the alignment of these models remains a pressing challenge as the policy tends to overfit the learned ``proxy" reward model past an inflection point of utility as measured by a ``gold" reward model that is more performant -- a phenomenon known as overoptimisation. Prior work has mitigated this issue by computing a pessimistic statistic over an ensemble of reward models, which is common in Offline Reinforcement Learning but incredibly costly for language models with high memory requirements, making such approaches infeasible for sufficiently large models. To this end, we propose using a shared encoder but separate linear heads. We find this leads to similar performance as the full ensemble while allowing tremendous savings in memory and time required for training for models of similar size. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2406_01013 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Scalable Ensembling For Mitigating Reward Overoptimisation Ahmed, Ahmed M. Rafailov, Rafael Sharkov, Stepan Li, Xuechen Koyejo, Sanmi Machine Learning Computation and Language Reinforcement Learning from Human Feedback (RLHF) has enabled significant advancements within language modeling for powerful, instruction-following models. However, the alignment of these models remains a pressing challenge as the policy tends to overfit the learned ``proxy" reward model past an inflection point of utility as measured by a ``gold" reward model that is more performant -- a phenomenon known as overoptimisation. Prior work has mitigated this issue by computing a pessimistic statistic over an ensemble of reward models, which is common in Offline Reinforcement Learning but incredibly costly for language models with high memory requirements, making such approaches infeasible for sufficiently large models. To this end, we propose using a shared encoder but separate linear heads. We find this leads to similar performance as the full ensemble while allowing tremendous savings in memory and time required for training for models of similar size. |
| title | Scalable Ensembling For Mitigating Reward Overoptimisation |
| topic | Machine Learning Computation and Language |
| url | https://arxiv.org/abs/2406.01013 |