LAB: Large-Scale Alignment for ChatBots
About
This work introduces LAB (Large-scale Alignment for chatBots), a novel methodology designed to overcome the scalability challenges in the instruction-tuning phase of large language model (LLM) training. Leveraging a taxonomy-guided synthetic data generation process and a multi-phase tuning framework, LAB significantly reduces reliance on expensive human annotations and proprietary models like GPT-4. We demonstrate that LAB-trained models can achieve competitive performance across several benchmarks compared to models trained with traditional human-annotated or GPT-4 generated synthetic data. Thus offering a scalable, cost-effective solution for enhancing LLM capabilities and instruction-following behaviors without the drawbacks of catastrophic forgetting, marking a step forward in the efficient training of LLMs for a wide range of applications.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Language Modeling and Reasoning | Open LLM Leaderboard | ARC81.6 | 33 | |
| Instruction Following | AlpacaEval GPT-4 (test) | AlpacaEval Win Rate (GPT-4)17.1 | 18 | |
| Question Answering | SQuAD 2.0 | BLEU-15.1 | 10 | |
| Question Answering Evaluation | Pop-QA Cities-20 | Factual Accuracy1.75 | 10 | |
| Question Answering Evaluation | SQuAD 2.0 | Factual Accuracy1.65 | 10 | |
| Question Answering | Pop-QA Cities-20 | BLEU-122.1 | 10 | |
| Question Answering | Wikitext-10 | BLEU-10.212 | 10 | |
| Question Answering Evaluation | Wikitext-10 | Factual Accuracy1.18 | 10 |