Papers › Finetuned Language Models Are Zero-Shot Learners

Finetuned Language Models Are Zero-Shot Learners

3 Sep 2021ICLR 2022 4arXiv:2109.01652archive 2025-07-28

Jason Wei, Maarten Bosma, Vincent Y. Zhao, Kelvin Guu, Adams Wei Yu, Brian Lester, Nan Du, Andrew M. Dai, Quoc V. Le

This paper explores a simple method for improving the zero-shot learning abilities of language models. We show that instruction tuning -- finetuning language models on a collection of tasks described via instructions -- substantially improves zero-shot performance on unseen tasks. We take a 137B parameter pretrained language model and instruction-tune it on over 60 NLP tasks verbalized via natural language instruction templates. We evaluate this instruction-tuned model, which we call FLAN, on unseen task types. FLAN substantially improves the performance of its unmodified counterpart and surpasses zero-shot 175B GPT-3 on 20 of 25 tasks that we evaluate. FLAN even outperforms few-shot GPT-3 by a large margin on ANLI, RTE, BoolQ, AI2-ARC, OpenbookQA, and StoryCloze. Ablation studies reveal that number of finetuning datasets, model scale, and natural language instructions are key to the success of instruction tuning.

PaperPDFConference PDFCodeCode Syntology ranMethod as code

In Syntology View this paper on Syntology, its page in Syntology's graph. That page adds which of its functions are also present in other papers' code Syntology holds, with dates, and its dated run record as JSON and a README badge. It also lists the repositories linked to the paper, the abstract and the calls for agents.

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

For agents, on Syntology's MCP service (how to connect):

Code

No sample from the archive's official repository (google-research/flan) is linked to this paper; the 1 sample below comes from 1 community repository. Syntology harvested code from google-research/flan for 2 other papers.

google-research/flan: Syntology's graph links these samples to Scaling Instruction-Finetuned Language Models (arXiv:2210.11416; that paper's own run record: 8 ran (of which 1 constructed an object rather than computing a result; 1 with no instrument failure: 0 honoured, 0 violated, 1 with no contract checked; 7 where Syntology's instrument failed) · 9 unverified) and Can Large Language Models Capture Dissenting Human Voices? (arXiv:2305.13788; that paper's own run record: 24 ran (of which 0 constructed an object rather than computing a result; 24 with no instrument failure: 0 honoured, 1 violated, 23 with no contract checked; 0 where Syntology's instrument failed) · 4 unverified); nothing checks that they implement this paper's method.

Syntology 1 code sample harvested from 1 repository linked to this paper: 0 ran · 1 unverified.

By repository: community (archive-listed): 1 sample from 1 repository, 0 ran (0 with no instrument failure; 0 where Syntology's instrument failed). The run record, sample by sample. “Ran” means executed on a synthesized input, not that the code is correct or reproduces the paper.

No sample is linked to this paper from 6 of its repositories that Syntology harvested for other papers:

google-research/flan officialmentioned in papermentioned on GitHubtfApache-2.0Syntology: no sample linked to this paper (harvested for another paper). Syntology's graph links these samples to Scaling Instruction-Finetuned Language Models (arXiv:2210.11416; that paper's own run record: 8 ran (of which 1 constructed an object rather than computing a result; 1 with no instrument failure: 0 honoured, 0 violated, 1 with no contract checked; 7 where Syntology's instrument failed) · 9 unverified) and Can Large Language Models Capture Dissenting Human Voices? (arXiv:2305.13788; that paper's own run record: 24 ran (of which 0 constructed an object rather than computing a result; 24 with no instrument failure: 0 honoured, 1 violated, 23 with no contract checked; 0 where Syntology's instrument failed) · 4 unverified); nothing checks that they implement this paper's method. report
bigcode-project/starcoder mentioned on GitHubpytorchSyntology: no sample linked to this paper (harvested for another paper). Syntology's graph links these samples to StarCoder: may the source be with you! (arXiv:2305.06161; that paper's own run record: 2 ran (of which 0 constructed an object rather than computing a result; 1 with no instrument failure: 1 honoured, 0 violated, 0 with no contract checked; 1 where Syntology's instrument failed) · 0 unverified); nothing checks that they implement this paper's method. report
bigscience-workshop/promptsource mentioned on GitHubSyntology: no sample linked to this paper (harvested for another paper). Syntology's graph links these samples to Multitask Prompted Training Enables Zero-Shot Task Generalization (arXiv:2110.08207; that paper's own run record: 13 ran (of which 0 constructed an object rather than computing a result; 9 with no instrument failure: 0 honoured, 1 violated, 8 with no contract checked; 4 where Syntology's instrument failed) · 2 unverified) and PromptSource: An Integrated Development Environment and Repository for Natural Language Prompts (arXiv:2202.01279; that paper's own run record: 1 ran (of which 1 constructed an object rather than computing a result; 1 with no instrument failure: 0 honoured, 0 violated, 1 with no contract checked; 0 where Syntology's instrument failed) · 0 unverified; the one sample that ran constructed an object rather than computing a result); nothing checks that they implement this paper's method. report
hiyouga/llama-efficient-tuning mentioned on GitHubpytorchApache-2.0Syntology: no sample linked to this paper (harvested for another paper). Syntology's graph links these samples to Large Language Models for Cyber Security: A Systematic Literature Review (arXiv:2405.04760; that paper's own run record: 3 ran (of which 0 constructed an object rather than computing a result; 1 with no instrument failure: 0 honoured, 0 violated, 1 with no contract checked; 2 where Syntology's instrument failed) · 0 unverified); nothing checks that they implement this paper's method. report
hojjat-mokhtarabadi/promptsource mentioned on GitHubApache-2.0Syntology: no sample linked to this paper (harvested for another paper). Syntology's graph links these samples to Multitask Prompted Training Enables Zero-Shot Task Generalization (arXiv:2110.08207; that paper's own run record: 13 ran (of which 0 constructed an object rather than computing a result; 9 with no instrument failure: 0 honoured, 1 violated, 8 with no contract checked; 4 where Syntology's instrument failed) · 2 unverified); nothing checks that they implement this paper's method. report
openbiolink/promptsource mentioned on GitHubApache-2.0Syntology: not harvested report
ukplab/arxiv2025-inherent-limits-plms mentioned on GitHubSyntology: 1 sample, 0 ran · 1 unverified report
MS-P3/code6 mindsporeSyntology: no sample linked to this paper (harvested for another paper). Syntology's graph links these samples to RAG and RAU: A Survey on Retrieval-Augmented Language Model in Natural Language Processing (arXiv:2404.19543; that paper's own run record: 1 ran (of which 0 constructed an object rather than computing a result; 1 with no instrument failure: 0 honoured, 0 violated, 1 with no contract checked; 0 where Syntology's instrument failed) · 0 unverified); nothing checks that they implement this paper's method. 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.

A paper named beside a repository with no sample linked to this paper is shown with that paper's own run record, not this paper's: “ran” means executed on a synthesized input, not that the code is correct, and “where Syntology's instrument failed” counts failures of Syntology's instrument, not of the code. Papers are listed in arXiv-id order, at most three per repository.

Code Syntology ran Syntology

1 sample harvested: 0 ran · 1 unverified. Syntology's record for this page has not changed since , the first build that kept a record date for it; when this build read Syntology's graph is in the build record. That date is not when the samples ran.

1unverified

Licence: 0 of the 1 sample are pointer only, meaning Syntology does not serve that copy's text. This page shows no code text for any sample; each one links to its file in the repository.

Harvested from ukplab/arxiv2025-inherent-limits-plms. “Ran” means the sample executed on a synthesized input. It does not mean the output is correct, and nothing here reproduces the paper's results. “Honoured” and “violated” refer to a contract Syntology drafted from the code itself. “Our draft was wrong” and “the contract check could not run” are failures of Syntology's instrument, not of the code: the sample ran on Syntology's synthesized input, and then the drafted contract did not describe it, or the check itself raised an error (RAISES), lacked a dependency (DEP_MISSING) or ran out of time (TIMEOUT); hover the label for which. A metamorphic tier, where shown, is evidence of the output's shape, not that the sample implements the method.

Each sample ends with its code_sha256. A sample's code_sha256 is Syntology's identity for that exact code. It is printed in full (16 hexadecimal characters for every sample in this build), and get_code(code_sha256="…") takes it exactly as printed. An agent calls it on Syntology's MCP service (how to connect); click an identity to copy that call.

Repository labels, per sample. official repository: The archive marks this repository official for the paper. named in the paper: The archive records that the paper mentions this repository; it is not marked official. community (archive-listed): In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper. found in paper text by Syntology: Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted. community: Not in the archive's code links for this paper; a community repository Syntology harvested. Samples are grouped by repository: a repository marked official first, then by how directly the repository is tied to the paper, then in the archive's order; never by how many ran. Licence labels name the repository's licence as recorded at harvest. “Pointer only” means Syntology does not serve that copy's text, for one of six reasons: no licence file was found; the licence was not identified; the licence is recorded as permissive but that copy's record is not marked cleared; the licence is outside the permissive list Syntology serves text under (MIT, Apache-2.0, BSD and similar); this paper's copy was not located, so its licence is not recorded; or this paper's copy is cleared, but the stored copy of the same code that Syntology serves is not. Some licences outside that list permit redistribution, such as WTFPL, and GPL-3.0 under its conditions; they are simply not on the list. Hover a licence label for the reason. File links open the file on GitHub at the default branch, which may have changed since the harvest.

ukplab/arxiv2025-inherent-limits-plms community (archive-listed) 1 sample, 0 ran · 1 unverified
preprocess_flan_dataset data_utils.py unverified Apache-2.0 (permissive) · 76f691da79652f18 · report

Method as code Syntology

What Syntology's graph records as this paper's claimed method, lifted into code: each assembly names its bricks (functions and classes), each with its own role and its own verification level. An assembly is not a check that the code implements the method correctly or reproduces the paper's results. Levels are each brick's own verification level in Syntology's graph (level 0: no recorded run), listed brick by brick and never combined into one level; “ran” means the brick executed on a synthesized input, not that its output is correct, and the instrument figure counts failures of Syntology's instrument, not of the code. Where a model read the code against the paper's claim, each assembly's label verdict is one model's judgement, kept in this page's JSON twin; it is not calibrated, so no card prints it.

1 assembly on this paper: 1 harvested, 0 generated.

harvested Assembly for instruction tuning (graph method: instruction tuning). 1 brick; individually on a fixture: 0 ran · 1 unverified. The assembly has not been run end-to-end.

  1. preprocess_flan_dataset entry · level 0 · unverified · 76f691da79652f18

Each brick ends with its code_sha256, Syntology's identity for that exact code; click one to copy the get_code call that returns it. For agents, on Syntology's MCP service: assembly(arxiv_id="2109.01652") returns these assemblies with each brick's role and wiring (how to connect).

Tasks

ARCCommon Sense ReasoningCoreference ResolutionLanguage ModelingLanguage ModellingMachine TranslationNatural Language InferenceQuestion AnsweringRTESentence CompletionSentiment AnalysisZero-Shot Learning

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Common Sense Reasoning ARC (Challenge) FLAN 137B (few-shot, k=13) Accuracy 63.8 #23 of 54 Archive leaderboard report
Common Sense Reasoning ARC (Challenge) FLAN 137B (zero-shot) Accuracy 63.1 #24 of 54 Archive leaderboard report
Common Sense Reasoning ARC (Easy) FLAN 137B (few-shot, k=14) Accuracy 80.7 #13 of 47 Archive leaderboard report
Common Sense Reasoning ARC (Easy) FLAN 137B (0-shot) Accuracy 79.6 #17 of 47 Archive leaderboard report
Common Sense Reasoning ReCoRD FLAN 137B (prompt-tuned) EM 85.1 #13 of 45 Archive leaderboard report
Common Sense Reasoning ReCoRD FLAN 137B (zero-shot) EM 72.5 #23 of 45 Archive leaderboard report
Common Sense Reasoning WinoGrande FLAN 137B (few-shot, k=16) Accuracy 72.8 #34 of 77 Archive leaderboard report
Common Sense Reasoning WinoGrande FLAN 137B (0-shot) Accuracy 71.2 #36 of 77 Archive leaderboard report
Coreference Resolution Winograd Schema Challenge FLAN 137B (prompt-tuned) Accuracy 86.5 #15 of 82 Archive leaderboard report
Coreference Resolution Winograd Schema Challenge FLAN 137B (zero-shot) Accuracy 80.8 #20 of 82 Archive leaderboard report
Machine Translation WMT2014 English-French FLAN 137B (zero-shot) BLEU score 33.9 #49 of 57 Archive leaderboard report
Machine Translation WMT2014 English-French FLAN 137B (few-shot, k=9) BLEU score 33.8 #50 of 57 Archive leaderboard report
Machine Translation WMT2014 French-English FLAN 137B (few-shot, k=9) BLEU score 37.9 #1 of 3 Archive leaderboard report
Machine Translation WMT2014 French-English FLAN 137B (zero-shot) BLEU score 35.9 #2 of 3 Archive leaderboard report
Machine Translation WMT2016 English-German FLAN 137B (zero-shot) BLEU score 27.0 #5 of 12 Archive leaderboard report
Machine Translation WMT2016 English-German FLAN 137B (few-shot, k=11) BLEU score 26.1 #7 of 12 Archive leaderboard report
Machine Translation WMT2016 English-Romanian FLAN 137B (few-shot, k=9) BLEU score 20.5 #19 of 21 Archive leaderboard report
Machine Translation WMT2016 English-Romanian FLAN 137B (zero-shot) BLEU score 18.9 #20 of 21 Archive leaderboard report
Machine Translation WMT2016 German-English FLAN 137B (few-shot, k=11) BLEU score 40.7 #1 of 8 Archive leaderboard report
Machine Translation WMT2016 German-English FLAN 137B (zero-shot) BLEU score 38.9 #2 of 8 Archive leaderboard report
Machine Translation WMT2016 Romanian-English FLAN 137B (few-shot, k=9) BLEU score 38.1 #2 of 21 Archive leaderboard report
Machine Translation WMT2016 Romanian-English FLAN 137B (zero-shot) BLEU score 37.3 #3 of 21 Archive leaderboard report
Natural Language Inference RTE FLAN 137B (prompt-tuned) Accuracy 91.7% #13 of 90 Archive leaderboard report
Natural Language Inference RTE FLAN 137B (8-shot) Accuracy 84.5% #29 of 90 Archive leaderboard report
Natural Language Inference RTE FLAN 137B (0-shot) Accuracy 84.1% #30 of 90 Archive leaderboard report
Natural Language Inference WNLI FLAN 137B (zero-shot) Accuracy 74.6 #14 of 23 Archive leaderboard report
Natural Language Inference WNLI FLAN 137B (few-shot, k=4) Accuracy 70.4 #17 of 23 Archive leaderboard report
Question Answering BoolQ FLAN 137B (prompt-tuned) Accuracy 86.3 #14 of 65 Archive leaderboard report
Question Answering BoolQ FLAN 137B (4-shot) Accuracy 84.6 #19 of 65 Archive leaderboard report
Question Answering BoolQ FLAN 137B (0-shot) Accuracy 82.9 #24 of 65 Archive leaderboard report
Question Answering COPA FLAN 137B (prompt-tuned) Accuracy 94 #10 of 60 Archive leaderboard report
Question Answering COPA FLAN 137B (zero-shot) Accuracy 91 #13 of 60 Archive leaderboard report
Question Answering COPA FLAN 137B (few-shot, k=16) Accuracy 87 #22 of 60 Archive leaderboard report
Question Answering MultiRC FLAN 137B (prompt-tuned) F1 83.4 #11 of 30 Archive leaderboard report
Question Answering MultiRC FLAN 137B (zero-shot) F1 77.5 #12 of 30 Archive leaderboard report
Question Answering MultiRC FLAN 137B (1-shot) F1 72.1 #14 of 30 Archive leaderboard report
Question Answering NaturalQA FLAN 137B (zero-shot) EM 20.7 #2 of 4 Archive leaderboard report
Question Answering OBQA FLAN 137B (zero-shot) Accuracy 78.4 #1 of 9 Archive leaderboard report
Question Answering OBQA FLAN 137B (few-shot, k=16) Accuracy 78.2 #2 of 9 Archive leaderboard report
Question Answering PIQA FLAN 137B (few-shot, k=10) Accuracy 81.7 #27 of 67 Archive leaderboard report
Question Answering PIQA FLAN 137B (0-shot) Accuracy 80.5 #31 of 67 Archive leaderboard report
Question Answering StoryCloze FLAN 137B (few-shot, k=10) Accuracy 94.7 #3 of 23 Archive leaderboard report
Question Answering StoryCloze FLAN 137B (zero-shot) Accuracy 93.4 #6 of 23 Archive leaderboard report
Question Answering TriviaQA FLAN 137B (zero-shot) EM 56.7 #42 of 56 Archive leaderboard report
Sentence Completion HellaSwag FLAN 137B (3-shot) Accuracy 59.2 #60 of 89 Archive leaderboard report
Sentence Completion HellaSwag FLAN 137B (0-shot) Accuracy 56.7 #62 of 89 Archive leaderboard report
Sentiment Analysis IMDb FLAN 137B (few-shot, k=2) Accuracy 95 #18 of 49 Archive leaderboard report
Sentiment Analysis IMDb FLAN 137B (zero-shot) Accuracy 94.3 #22 of 49 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

AdamAttentionAttention DropoutBPECosine AnnealingDense ConnectionsDropoutGPT-3Layer NormalizationLinear LayerLinear Warmup With Cosine AnnealingMulti-Head AttentionResidual ConnectionSoftmaxWeight Decay

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