Papers › The CoT Collection: Improving Zero-shot and Few-shot Learning of Language Models via...

The CoT Collection: Improving Zero-shot and Few-shot Learning of Language Models via Chain-of-Thought Fine-Tuning

23 May 2023arXiv:2305.14045archive 2025-07-28

Seungone Kim, Se June Joo, Doyoung Kim, Joel Jang, Seonghyeon Ye, Jamin Shin, Minjoon Seo

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.

PaperPDFCode

In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

kaist-lklab/cot-collection officialmentioned in papermentioned on GitHubpytorch report
kaistai/cot-collection officialmentioned in papermentioned on GitHubpytorch report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Common Sense ReasoningCommon Sense Reasoning (Zero-Shot)Coreference ResolutionFew-Shot LearningNatural Language InferenceNatural Language Inference (Zero-Shot)Question AnsweringSentence CompletionWord Sense Disambiguation

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
BIG-bench (Hyperbaton) CoT-T5 11B Accuracy 65.2 #1 of 1 Archive leaderboard report
BIG-bench (Navigate) CoT-T5 11B Accuracy 60 #1 of 1 Archive leaderboard report
BIG-bench (Ruin Names) CoT-T5 11B Accuracy 42.8 #1 of 1 Archive leaderboard report
BIG-bench (SNARKS) CoT-T5 11B Accuracy 67.7 #1 of 1 Archive leaderboard report
Big-bench Hard CoT-T5 11B Accuracy 48 #1 of 1 Archive leaderboard report
Common Sense Reasoning WinoGrande T0-3B (CoT fine-tuned) Accuracy 57.5 #57 of 77 Archive leaderboard report
Coreference Resolution Winograd Schema Challenge T0-3B (CoT fine-tuned) Accuracy 66 #41 of 82 Archive leaderboard report
Few-Shot Learning CaseHOLD CoT-T5-11B (1024 Shot) Accuracy 68.3 #1 of 1 Archive leaderboard report
Few-Shot Learning MedNLI CoT-T5-11B (1024 Shot) Accuracy 78.02 #1 of 1 Archive leaderboard report
Few-Shot Learning PubMedQA CoT-T5-11B (1024 Shot) Accuracy 73.42 #2 of 2 Archive leaderboard report
Natural Language Inference ANLI test T0-3B (CoT fine-tuned) A1 41.7 #11 of 25 Archive leaderboard report
Natural Language Inference ANLI test T0-3B (CoT fine-tuned) A2 37.2 #11 of 25 Archive leaderboard report
Natural Language Inference ANLI test T0-3B (CoT fine-tuned) A3 41.9 #11 of 25 Archive leaderboard report
Natural Language Inference RTE T0-3B (CoT fine-tuned) Accuracy 80.8% #34 of 90 Archive leaderboard report
Question Answering COPA T0-3B (CoT fine-tuned) Accuracy 90.9 #16 of 60 Archive leaderboard report
Question Answering PubMedQA CoT-T5-11B (1024 Shot) Accuracy 73.42 #21 of 30 Archive leaderboard report
Question Answering StoryCloze T0-3B (CoT fine-tuned) Accuracy 94.5 #4 of 23 Archive leaderboard report
Sentence Completion HellaSwag T0-3B (CoT fine-tuned) Accuracy 41.1 #75 of 89 Archive leaderboard report
Word Sense Disambiguation Words in Context T0-3B (CoT fine-tuned) Accuracy 56.7 #21 of 37 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

Flan-T5

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections