Inference Time Alignment with Reward-Guided Tree Search
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_ | 1866916495576006656 |
|---|---|
| author | Hung, Chia-Yu Majumder, Navonil Mehrish, Ambuj Poria, Soujanya |
| author_facet | Hung, Chia-Yu Majumder, Navonil Mehrish, Ambuj Poria, Soujanya |
| contents | Inference-time computation methods enhance the performance of Large Language Models (LLMs) by leveraging additional computational resources to achieve superior results. Common techniques, such as Best-of-N sampling, Majority Voting, and variants of tree-search algorithms have proven to be effective in boosting the performance of LLMs. These approaches strategically trade increased computational resources for improved model responses. In this work, we proposed DARWIN, an inference-time alignment method that leverages the guidance of a reward model to achieve alignment through a reward-guided tree search. Empirical evidences indicates that our method outperforms other inference-time alignment methods such as Best-of-N and ARGS on two widely accepted alignment benchmarks AlpacaEval 2 and MT-Bench. Furthermore, we show that our inference-time approach achieves performance comparable to preference-tuned models on both benchmarks, highlighting the effectiveness of trading inference-time compute for enhanced performance during inference. We have released our codes at https://github.com/declare-lab/darwin. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2406_15193 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Inference Time Alignment with Reward-Guided Tree Search Hung, Chia-Yu Majumder, Navonil Mehrish, Ambuj Poria, Soujanya Computation and Language Inference-time computation methods enhance the performance of Large Language Models (LLMs) by leveraging additional computational resources to achieve superior results. Common techniques, such as Best-of-N sampling, Majority Voting, and variants of tree-search algorithms have proven to be effective in boosting the performance of LLMs. These approaches strategically trade increased computational resources for improved model responses. In this work, we proposed DARWIN, an inference-time alignment method that leverages the guidance of a reward model to achieve alignment through a reward-guided tree search. Empirical evidences indicates that our method outperforms other inference-time alignment methods such as Best-of-N and ARGS on two widely accepted alignment benchmarks AlpacaEval 2 and MT-Bench. Furthermore, we show that our inference-time approach achieves performance comparable to preference-tuned models on both benchmarks, highlighting the effectiveness of trading inference-time compute for enhanced performance during inference. We have released our codes at https://github.com/declare-lab/darwin. |
| title | Inference Time Alignment with Reward-Guided Tree Search |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2406.15193 |