Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2412.04305 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866929616943316992 |
|---|---|
| author | Yasunaga, Michihiro Shamis, Leonid Zhou, Chunting Cohen, Andrew Weston, Jason Zettlemoyer, Luke Ghazvininejad, Marjan |
| author_facet | Yasunaga, Michihiro Shamis, Leonid Zhou, Chunting Cohen, Andrew Weston, Jason Zettlemoyer, Luke Ghazvininejad, Marjan |
| contents | Recent approaches to large language model (LLM) alignment typically require millions of human annotations or rely on external aligned models for synthetic data generation. This paper introduces ALMA: Alignment with Minimal Annotation, demonstrating that effective alignment can be achieved using only 9,000 labeled examples -- less than 1% of conventional approaches. ALMA generates large amounts of high-quality synthetic alignment data through new techniques: diverse prompt synthesis via few-shot learning, diverse response generation with multiple model checkpoints, and judge (reward model) enhancement through score aggregation and self-distillation. Using only a pretrained Llama3 base model, 5,000 SFT examples, and 4,000 judge annotations, ALMA achieves performance close to Llama3-Instruct across diverse alignment benchmarks (e.g., 0.1% difference on AlpacaEval 2.0 score). These results are achieved with a multi-round, self-bootstrapped data synthesis and training recipe that continues to improve for 10 rounds, surpassing the typical 3-round ceiling of previous methods. These results suggest that base models already possess sufficient knowledge for effective alignment, and that synthetic data generation methods can expose it. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2412_04305 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | ALMA: Alignment with Minimal Annotation Yasunaga, Michihiro Shamis, Leonid Zhou, Chunting Cohen, Andrew Weston, Jason Zettlemoyer, Luke Ghazvininejad, Marjan Computation and Language Machine Learning Recent approaches to large language model (LLM) alignment typically require millions of human annotations or rely on external aligned models for synthetic data generation. This paper introduces ALMA: Alignment with Minimal Annotation, demonstrating that effective alignment can be achieved using only 9,000 labeled examples -- less than 1% of conventional approaches. ALMA generates large amounts of high-quality synthetic alignment data through new techniques: diverse prompt synthesis via few-shot learning, diverse response generation with multiple model checkpoints, and judge (reward model) enhancement through score aggregation and self-distillation. Using only a pretrained Llama3 base model, 5,000 SFT examples, and 4,000 judge annotations, ALMA achieves performance close to Llama3-Instruct across diverse alignment benchmarks (e.g., 0.1% difference on AlpacaEval 2.0 score). These results are achieved with a multi-round, self-bootstrapped data synthesis and training recipe that continues to improve for 10 rounds, surpassing the typical 3-round ceiling of previous methods. These results suggest that base models already possess sufficient knowledge for effective alignment, and that synthetic data generation methods can expose it. |
| title | ALMA: Alignment with Minimal Annotation |
| topic | Computation and Language Machine Learning |
| url | https://arxiv.org/abs/2412.04305 |