Command Palette
Search for a command to run...
The CoT Collection: Improving Zero-shot and Few-shot Learning of Language Models via Chain-of-Thought Fine-Tuning
Seungone Kim Se June Joo Doyoung Kim Joel Jang Seonghyeon Ye Jamin Shin Minjoon Seo
Abstract
Language models (LMs) with less than 100B parameters are known to perform poorly on chain-of-thought (CoT) reasoning in contrast to large LMs when solving unseen tasks. In this work, we aim to equip smaller LMs with the step-by-step reasoning capability by instruction tuning with CoT rationales. In order to achieve this goal, we first introduce a new instruction-tuning dataset called the CoT Collection, which augments the existing Flan Collection (including only 9 CoT tasks) with additional 1.84 million rationales across 1,060 tasks. We show that CoT fine-tuning Flan-T5 (3B & 11B) with CoT Collection enables smaller LMs to have better CoT capabilities on unseen tasks. On the BIG-Bench-Hard (BBH) benchmark, we report an average improvement of +4.34% (Flan-T5 3B) and +2.60% (Flan-T5 11B), in terms of zero-shot task accuracy. Furthermore, we show that instruction tuning with CoT Collection allows LMs to possess stronger few-shot learning capabilities on 4 domain-specific tasks, resulting in an improvement of +2.24% (Flan-T5 3B) and +2.37% (Flan-T5 11B), even outperforming ChatGPT utilizing demonstrations until the max length by a +13.98% margin. Our code, the CoT Collection data, and model checkpoints are publicly available.
Code Repositories
Benchmarks
| Benchmark | Methodology | Metrics |
|---|---|---|
| common-sense-reasoning-on-winogrande | T0-3B (CoT fine-tuned) | Accuracy: 57.5 |
| coreference-resolution-on-winograd-schema | T0-3B (CoT fine-tuned) | Accuracy: 66 |
| few-shot-learning-on-casehold | CoT-T5-11B (1024 Shot) | Accuracy: 68.3 |
| few-shot-learning-on-mednli | CoT-T5-11B (1024 Shot) | Accuracy: 78.02 |
| few-shot-learning-on-pubmedqa | CoT-T5-11B (1024 Shot) | Accuracy: 73.42 |
| natural-language-inference-on-anli-test | T0-3B (CoT fine-tuned) | A1: 41.7 A2: 37.2 A3: 41.9 |
| natural-language-inference-on-rte | T0-3B (CoT fine-tuned) | Accuracy: 80.8% |
| question-answering-on-copa | T0-3B (CoT fine-tuned) | Accuracy: 90.9 |
| question-answering-on-pubmedqa | CoT-T5-11B (1024 Shot) | Accuracy: 73.42 |
| question-answering-on-storycloze | T0-3B (CoT fine-tuned) | Accuracy: 94.5 |
| word-sense-disambiguation-on-words-in-context | T0-3B (CoT fine-tuned) | Accuracy: 56.7 |
Build AI with AI
From idea to launch — accelerate your AI development with free AI co-coding, out-of-the-box environment and best price of GPUs.