decision-mix-large

Training data for a decision model: a state (text, or an object with named parts) plus 1–4 questions about it, answered as a yes/no probability (noul), a pick from named options (choice) or a level on an ordered scale (score). Rows come from public datasets (Hugging Face Hub, plus a few on GitHub) that allow commercial use, labeled by people, votes, rules, tests or a simulator (no LLM labels). States are in English, Arabic, Chinese, French, German, Italian, Japanese, Portuguese, Spanish, Dutch, Polish, Russian, Turkish, Vietnamese or Hindi.

Splits

Split	Rows	Question mix (noul/choice/score)	Questions per example (1/2/3/4)
train	9,749,485	43%/38%/19%	48%/47%/4%/1%
test	515,775	43%/38%/19%	48%/47%/4%/1%
test_heldout_tasks	5,286	47%/35%/18%	40%/60%/0%/0%
calibration	3,000	43%/39%/18%	53%/43%/3%/1%

• test: 5% of each training task, with no state shared with train. It comes from the source's own test split when there is one, topped up with source rows kept out of train when that split is too small.
• test_heldout_tasks: 4 tasks never used in training (cefr_sentence_level, financial_tweet_topic, mt_bench_human_winner, support_urgency).
• calibration: rows from the training tasks, disjoint from train and test.

Format

Each row has four columns: example (a JSON string), task (the source task), n_questions and n_chars (the length of example in characters, handy for length bucketing or filtering). A decoded example is a state plus 1–4 questions about it:

{
  "state": "Hi, I was charged twice for order #4411 and need it fixed today.",
  "questions": {
    "urgent":   {"type": "noul", "instructions": "Does this convey urgency?", "label": 1.0},
    "intent":   {"type": "choice", "instructions": "What does the customer want?",
                 "criteria": {"refund": "money back for a charge", "bug": null, "other": null}, "label": "refund"},
    "severity": {"type": "score", "instructions": "How severe is the issue?",
                 "criteria": ["low", "medium", "high"], "label": "high"}
  }
}

• state: a string, an object (serialize with sorted keys) or a list of strings (join with newlines). Instructions about an object state name the part they mean, e.g. `hypothesis_2`.
• noul: no criteria; label is a float in [0, 1]. Soft values appear only in tasks where the label is the share of annotators who said yes.
• choice: criteria is {name: description or null} or a list of names (2–255, unique); label is one name. Options are shuffled so the correct position is uniform.
• score: criteria lists the levels from lowest to highest (never reordered); label is one level.

Loading

import json
from datasets import load_dataset

ds = load_dataset("paulprt/decision-mix-large")
row = ds["train"][0]
example = json.loads(row["example"])

How it was built

• Balance (per task and split): noul questions are 45–55% positive, no score level exceeds 40%, and the correct choice option is uniform across positions. Imbalance is fixed by dropping majority-class questions, never by duplicating. Groups under 100 questions may miss a bound by one example.
• Caps and mix: at most 150,000 train examples per task. Task weights are tuned towards a 40/40/20 noul/choice/score mix, within 2× of their base weight. At this size most tasks use their whole balanced pool and every available score question is used, so the train mix is the one in the table above; reweight question types at training time if needed.
• Grouping: about half the examples ask one question and half ask 2–4 about the same state. Groups are natural ones (hypotheses of one premise, questions on one passage, products for one query, translations of one source) or checks that follow from another question's label ("is it at least X?"); a check is never asked on its own.
• Instructions: at least 8 phrasings per question. Classification sets with many labels sometimes show every label and sometimes the answer plus 3–10 distractors.
• Limits: every example (state plus questions and options) fits in 2,048 LFM2.5-Encoder-230M tokens, the questions alone in 1,024. No state appears twice in the dataset (normalised text hash).
• Determinism: the build is deterministic (seed 0). stats.md in this repo has per-task and per-split counts, label balance, position histograms, score distributions and token percentiles. meta/tasks.json lists each task's sources, license and counts.

License

Every row keeps the license of the dataset it came from (see the table). All sources allow commercial use. Rows from share-alike sources (CC-BY-SA, CDLA-Sharing) must keep those terms if redistributed. cmedqa2_answer_match, dynahate_hate, phishing_email, tweets_hate_speech, twitteremo_polish_emotions come from GPL or LGPL sources (copyleft): keep those terms if you redistribute their rows. multilingual_toxicity, paradetox_less_toxic are under OpenRAIL++, which adds use-based restrictions. Please credit the source datasets.

Tasks

Task	Domain	Source	License	Labels	Train	Test	Calibration	Held-out
mind2web_next_action	agents	osunlp/Mind2Web	cc-by-4.0	human	6,350	334	2	0
mtop_tool_calls	agents	tasksource/mtop	cc-by-sa-4.0	rules	22,770	1,198	7	0
swe_tool_outcome	agents	nebius/SWE-agent-trajectories	cc-by-4.0	rules	24,024	1,264	7	0
webchain_next_action	agents	cklxx/webchain-elements	cc-by-4.0	human	7,045	371	2	0
bird_sql_match	code	birdsql/bird23-train-filtered	cc-by-sa-4.0	human	6,056	319	2	0
code_contests_plus_correctness	code	ByteDance-Seed/Code-Contests-Plus	cc-by-4.0	tests	11,995	631	4	0
code_language_id	code	codeparrot/github-code-clean	apache-2.0	rules	37,656	1,982	11	0
codecomplex_time_complexity	code	codeparrot/codecomplex	apache-2.0	human	3,167	167	1	0
codeforces_submission_verdict	code	open-r1/codeforces-submissions open-r1/codeforces	cc-by-4.0	tests	30,186	1,589	9	0
codenet_same_problem	code	qiankunmu/Project_CodeNet_Python800_and_Java250	apache-2.0	rules	12,109	637	4	0
codet_m4_generator	code	DaniilOr/CoDET-M4	mit	rules	65,340	3,439	20	0
codet_m4_machine_code	code	DaniilOr/CoDET-M4	mit	rules	46,287	2,700	17	0
commit_message_match	code	bigcode/commitpackft	mit	rules	40,236	2,118	12	0
cosqa_query_code_match	code	gonglinyuan/CoSQA	mit	human	12,235	644	4	0
cubert_docstring_match	code	claudios/cubert_ETHPy150Open	apache-2.0	rules	46,216	2,432	14	0
cubert_exception_type	code	claudios/cubert_ETHPy150Open	apache-2.0	rules	17,340	913	5	0
cubert_swapped_operands	code	claudios/cubert_ETHPy150Open	apache-2.0	rules	42,227	2,225	13	0
cubert_variable_misuse	code	claudios/cubert_ETHPy150Open	apache-2.0	rules	47,931	2,540	15	0
cubert_wrong_binary_operator	code	claudios/cubert_ETHPy150Open	apache-2.0	rules	39,014	2,248	13	0
cve_severity	code	CIRCL/vulnerability-scores	cc-by-4.0	human	114,906	6,048	34	0
github_issue_type	code	sharjeelyunus/github-issues-dataset	mit	human	34,400	1,811	10	0
h_airosetta_machine_code	code	isThisYouLLM/H-AIRosettaMP	mit	rules	18,653	982	7	0
primevul_vulnerable_function	code	colin/PrimeVul	mit	rules	5,767	304	2	0
pypi_development_status	code	open-index/open-pypi	odc-by	human	99,064	5,215	30	0
pypi_intended_audience	code	open-index/open-pypi	odc-by	human	7,058	372	9	0
pypi_topic	code	open-index/open-pypi	odc-by	human	5,820	310	8	0
spider_sql_match	code	xlangai/spider	cc-by-sa-4.0	human	6,779	357	2	0
stackoverflow_question_attention	code	pacovaldez/stackoverflow-questions	apache-2.0	votes	150,000	7,895	45	0
swe_patch_resolves_issue	code	nebius/SWE-rebench-openhands-trajectories nebius/SWE-agent-trajectories nebius/SWE-rebench nebius/SWE-bench-extra	cc-by-4.0	tests	9,884	520	3	0
taco_problem_difficulty	code	BAAI/TACO	apache-2.0	human	5,986	315	2	0
attributionbench_claims	grounding	osunlp/AttributionBench	apache-2.0	human	8,986	473	3	0
boolq_yes_no	grounding	google/boolq	cc-by-sa-3.0	human	7,655	403	2	0
circa_indirect_answers	grounding	google-research-datasets/circa	cc-by-4.0	human	19,322	1,017	6	0
claimbuster_checkworthiness	grounding	Nithiwat/claimbuster	cc-by-sa-4.0	human	14,935	786	4	0
cmedqa2_answer_match	grounding	fzkuji/cMedQA2	gpl-3.0	rules	28,075	1,478	8	0
cmrc2018_answer_check	grounding	hfl/cmrc2018	cc-by-sa-4.0	human	4,194	221	1	0
code_contests_correctness	grounding	deepmind/code_contests	cc-by-4.0	tests	6,015	317	2	0
defeasible_nli_update	grounding	tasksource/defeasible-nli	apache-2.0	human	29,677	1,562	9	0
drop_numeric_reasoning	grounding	ucinlp/drop	cc-by-sa-4.0	human	29,063	1,530	9	0
duorc_plot_answerability	grounding	ibm-research/duorc	mit	human	10,257	541	3	0
esci_product_relevance	grounding	tasksource/esci	apache-2.0	human	45,785	2,408	14	0
expertqa_claim_judgments	grounding	cmalaviya/expertqa	mit	human	8,356	440	3	0
faithdial_hallucination	grounding	McGill-NLP/FaithDial	mit	human	11,437	602	3	0
fever_claim_verification	grounding	copenlu/fever_gold_evidence	cc-by-sa-3.0	human	66,219	3,485	20	0
finqa_numeric_answer	grounding	bevaya/FinQA	mit	human	6,918	364	2	0
fool_me_twice_verification	grounding	tasksource/fool-me-twice	apache-2.0	human	11,208	590	3	0
hagrid_attribution	grounding	miracl/hagrid	apache-2.0	human	2,156	113	1	0
headline_cause_relation	grounding	IlyaGusev/headline_cause	cc0-1.0	human	11,587	610	3	0
hotpotqa_supporting_paragraphs	grounding	hotpotqa/hotpot_qa	cc-by-sa-4.0	human	22,580	1,190	7	0
hybridqa_table_text	grounding	bentrevett/hybrid_qa	cc-by-4.0	human	33,672	1,772	10	0
jglue_jnli	grounding	github:yahoojapan/JGLUE	cc-by-sa-4.0	human	16,161	867	5	0
klej_dyk_answer_check	grounding	allegro/klej-dyk	cc-by-sa-3.0	human	933	49	0	0
liar2_truthfulness	grounding	chengxuphd/liar2	apache-2.0	human	17,798	937	5	0
miracl_passage_relevance	grounding	miracl/miracl miracl/miracl-corpus	apache-2.0	human	5,535	292	2	0
mnli_entailment	grounding	nyu-mll/multi_nli	cc-by-3.0	human	63,563	3,345	19	0
multihiertt_financial_tables	grounding	yilunzhao/MultiHiertt	mit	human	6,506	342	2	0
piaf_answerability	grounding	AgentPublic/piaf	mit	rules	6,603	348	2	0
polqa_passage_relevance	grounding	ipipan/polqa	cc-by-sa-4.0	human	13,531	713	4	0
poquad_answer_check	grounding	clarin-pl/poquad	cc-by-4.0	human	12,968	683	4	0
pubmedqa_decision	grounding	qiaojin/PubMedQA	mit	heuristic	26,617	1,401	8	0
qe4pe_translation_quality	grounding	gsarti/qe4pe	apache-2.0	human	2,157	114	1	0
rsg_inference_mix	grounding	RussianNLP/russian_super_glue	mit	human	5,410	285	2	0
rsg_muserc_answers	grounding	RussianNLP/russian_super_glue	mit	human	3,953	208	1	0
rsg_rucos_cloze	grounding	RussianNLP/russian_super_glue	mit	rules	37,334	1,965	11	0
scinli_relation	grounding	tasksource/scinli	apache-2.0	rules	44,177	2,325	13	0
seahorse_summary_quality	grounding	tasksource/seahorse_summarization_evaluation	cc-by-4.0	human	22,292	1,173	7	0
snli_entailment	grounding	stanfordnlp/snli	cc-by-sa-4.0	human	68,938	3,628	21	0
squad2_answerability	grounding	rajpurkar/squad_v2	cc-by-sa-4.0	human	10,831	570	3	0
t2ranking_passage_relevance	grounding	THUIR/T2Ranking	apache-2.0	human	96,822	5,096	29	0
tabfact_verification	grounding	github:wenhuchen/Table-Fact-Checking	mit	human	14,753	777	4	0
tatqa_financial_tables	grounding	next-tat/TAT-QA	cc-by-4.0	human	2,139	113	1	0
tydiqa_passage_answer	grounding	google-research-datasets/tydiqa	apache-2.0	human	15,053	793	5	0
vitaminc_fact_verification	grounding	tals/vitaminc	cc-by-sa-3.0	human	69,114	3,638	21	0
viwikifc_claim_verification	grounding	High-Will/ViWikiFC	cc-by-4.0	human	16,212	853	5	0
wands_product_relevance	grounding	tasksource/wands	mit	human	44,961	2,368	14	0
wice_citation_support	grounding	tasksource/wice	cc-by-sa-4.0	human	3,615	190	1	0
wiki_hades_hallucination	grounding	tasksource/wiki-hades	apache-2.0	human	7,771	409	2	0
wikitablequestions_answer	grounding	github:ppasupat/WikiTableQuestions	cc-by-sa-4.0	human	11,574	609	3	0
wmt_da_translation_quality	grounding	RicardoRei/wmt-da-human-evaluation	apache-2.0	human	43,410	2,283	13	0
wmt_mqm_translation_errors	grounding	RicardoRei/wmt-mqm-human-evaluation	apache-2.0	human	9,016	475	3	0
wmt_sqm_translation_quality	grounding	RicardoRei/wmt-sqm-human-evaluation	apache-2.0	human	2,527	134	1	0
xfact_verdict	grounding	utahnlp/x-fact	mit	human	11,528	607	3	0
acl_paper_citation_quartile	judging	nhop/scientific-quality-score-prediction	cc-by-4.0	votes	38,297	2,016	12	0
arena_140k_winner	judging	lmarena-ai/arena-human-preference-140k	cc-by-4.0	votes	60,338	3,176	18	0
arena_55k_winner	judging	lmarena-ai/arena-human-preference-55k	apache-2.0	votes	47,732	2,513	14	0
arxiv_citation_quartile	judging	yufan/arxiv-metadata-2020-2026	odc-by	votes	150,000	7,895	45	0
chinese_herbal_product_review_stars	judging	OpenModels/Chinese-Herbal-Medicine-Sentiment	mit	votes	83,384	4,389	26	0
community_note_helpful	judging	histlearn/community-notes-br	cdla-permissive-2.0	votes	48,813	2,569	15	0
feedbackqa_answer_rating	judging	McGill-NLP/feedbackQA	apache-2.0	human	4,199	221	1	0
helpsteer2_attributes	judging	nvidia/HelpSteer2	cc-by-4.0	human	17,477	920	5	0
helpsteer3_edit_quality	judging	nvidia/HelpSteer3	cc-by-4.0	human	1,967	104	1	0
helpsteer3_preference	judging	nvidia/HelpSteer3	cc-by-4.0	human	19,756	1,040	6	0
helpsteer_attributes_v1	judging	nvidia/HelpSteer	cc-by-4.0	human	30,892	1,626	9	0
hh_rlhf_harmless	judging	Anthropic/hh-rlhf	mit	human	41,064	2,161	12	0
hh_rlhf_helpful	judging	Anthropic/hh-rlhf	mit	human	71,245	3,750	21	0
instruction_constraint_checks	judging	OpenAssistant/oasst2	apache-2.0	rules	37,853	1,991	11	0
it_subreddit_comment_score	judging	NetGene/reddit-it-labor-sentiment-2020-2026	mit	votes	150,000	7,895	45	0
metacritic_review_spoiler	judging	Wada1/Metacritic_Games_Reviews_Dataset	cc-by-4.0	human	20,769	1,094	6	0
metacritic_user_review_score	judging	Wada1/Metacritic_Games_Reviews_Dataset	cc-by-4.0	votes	137,159	7,170	45	0
mt_bench_human_winner	judging	lmsys/mt_bench_human_judgments	cc-by-4.0	human	0	0	0	1,500
multipref_preference	judging	allenai/multipref	odc-by	human	7,623	401	2	0
nigerian_bank_app_review_stars	judging	Federal-University-Lokoja/Bank-Review	apache-2.0	votes	149,906	7,894	45	0
oasst2_message_labels	judging	OpenAssistant/oasst2	apache-2.0	human	70,146	3,692	21	0
oasst2_reply_ranking	judging	OpenAssistant/oasst2	apache-2.0	human	24,654	1,297	7	0
openreview_review_score	judging	nhop/scientific-quality-score-prediction	cc-by-4.0	human	26,560	1,401	8	0
prm800k_step_rating	judging	tasksource/PRM800K	mit	human	34,909	1,839	11	0
reddit_reply_upvotes	judging	allenai/compred	odc-by	votes	40,711	2,143	12	0
saudi_restaurant_review_stars	judging	Musaed1/kudu-google-maps-reviews	apache-2.0	votes	90,980	4,793	27	0
stackexchange_answer_preference	judging	HuggingFaceH4/stack-exchange-preferences	cc-by-sa-4.0	votes	87,249	4,592	26	0
zhihu_answer_upvotes	judging	liyucheng/zhihu_26k	cc-by-2.0	votes	21,364	1,124	6	0
ambigqa_ambiguity	qa	sewon/ambig_qa	cc-by-sa-3.0	human	4,445	234	1	0
aqua_rat_math	qa	deepmind/aqua_rat	apache-2.0	human	19,142	1,015	6	0
arc_science	qa	allenai/ai2_arc	cc-by-sa-4.0	human	3,244	171	1	0
balanced_copa	qa	pkavumba/balanced-copa	cc-by-4.0	human	966	51	0	0
commonsense_qa	qa	tau/commonsense_qa	mit	human	6,803	358	2	0
csqa2_assertion	qa	tasksource/commonsense_qa_2.0	cc-by-4.0	human	4,269	225	1	0
jglue_jcommonsenseqa	qa	github:yahoojapan/JGLUE	cc-by-sa-4.0	human	6,492	342	2	0
mathqa	qa	regisss/math_qa	apache-2.0	human	9,024	475	4	0
medmcqa	qa	openlifescienceai/medmcqa	apache-2.0	human	22,406	1,179	7	0
medqa_usmle	qa	GBaker/MedQA-USMLE-4-options	cc-by-4.0	human	6,972	367	2	0
narrativeqa_answer_check	qa	deepmind/narrativeqa	apache-2.0	human	4,683	246	1	0
qasc_science	qa	allenai/qasc	cc-by-4.0	human	6,033	318	2	0
qasper_evidence	qa	allenai/qasper	cc-by-4.0	human	2,016	106	1	0
quartz_qualitative	qa	allenai/quartz	cc-by-4.0	human	2,971	156	1	0
ropes_situation_answer	qa	allenai/ropes	cc-by-4.0	human	7,209	379	2	0
selqa_answer_sentence	qa	community-datasets/selqa	apache-2.0	human	3,153	166	1	0
selqa_question_topic	qa	community-datasets/selqa	apache-2.0	human	5,503	290	2	0
aegis1_response_safety	safety	nvidia/Aegis-AI-Content-Safety-Dataset-1.0	cc-by-4.0	human	5,629	296	2	0
aegis2_safety	safety	nvidia/Aegis-AI-Content-Safety-Dataset-2.0	cc-by-4.0	human	15,330	811	5	0
aya_redteam_harm_categories	safety	CohereLabs/aya_redteaming	apache-2.0	human	3,631	191	1	0
civil_comments_toxicity	safety	google/civil_comments	cc0-1.0	votes	54,228	2,854	16	0
cold_chinese_offensive	safety	thu-coai/cold	apache-2.0	human	17,047	897	5	0
deepset_prompt_injection	safety	deepset/prompt-injections	apache-2.0	human	419	22	0	0
diasafety_reply_safety	safety	thu-coai/diasafety	apache-2.0	human	9,779	515	3	0
dynahate_hate	safety	tasksource/dynahate	gpl	human	20,478	1,077	6	0
edos_sexism	safety	tasksource/edos	cc0-1.0	human	7,791	410	2	0
ethics_commonsense	safety	hendrycks/ethics	mit	human	8,793	463	3	0
ethics_deontology	safety	hendrycks/ethics	mit	human	10,050	529	3	0
ethics_justice	safety	hendrycks/ethics	mit	human	11,006	579	3	0
ethics_utilitarianism	safety	hendrycks/ethics	mit	human	13,256	698	4	0
ethics_virtue	safety	hendrycks/ethics	mit	human	5,197	274	2	0
germeval18_offensive	safety	jfrei/GermEval18	cc-by-4.0	human	4,672	247	1	0
gretel_pii_types	safety	gretelai/gretel-pii-masking-en-v1	apache-2.0	rules	21,870	1,151	7	0
hh_red_team_success	safety	Anthropic/hh-rlhf	mit	human	34,255	1,803	10	0
jailbreak_in_the_wild	safety	TrustAIRLab/in-the-wild-jailbreak-prompts	mit	human	1,762	93	1	0
jigsaw_toxic_comments	safety	tasksource/jigsaw_toxicity	apache-2.0	human	15,491	815	5	0
llmail_injection_outcome	safety	microsoft/llmail-inject-challenge	mit	simulator	22,326	1,175	7	0
measuring_hate_speech	safety	ucberkeley-dlab/measuring-hate-speech	cc-by-4.0	human	30,140	1,589	9	0
moral_stories_norm	safety	demelin/moral_stories	mit	human	10,718	564	3	0
multilingual_toxicity	safety	textdetox/multilingual_toxicity_dataset	openrail++	human	21,247	1,188	7	0
nemotron_multilingual_safety	safety	nvidia/Nemotron-Safety-Guard-Dataset-v3	cc-by-4.0	human	42,869	2,254	13	0
neuralchemy_prompt_attacks	safety	neuralchemy/Prompt-injection-dataset	apache-2.0	human	3,648	192	1	0
nuha_online_gender_violence	safety	thejosango/nuha-ajp-dataset	apache-2.0	human	31,927	1,681	10	0
paradetox_less_toxic	safety	textdetox/multilingual_paradetox	openrail++	human	2,508	132	1	0
phreshphish_phishing_page	safety	phreshphish/phreshphish	cc-by-4.0	rules	22,018	1,159	7	0
polish_cyberbullying	safety	ptaszynski/PolishCyberbullyingDataset	cc-by-4.0	human	2,964	156	1	0
prosocial_dialog_safety	safety	allenai/prosocial-dialog	cc-by-4.0	human	114,205	6,021	34	0
russian_inappropriate_message	safety	NiGuLa/Russian_Inappropriate_Messages	apache-2.0	human	27,678	1,457	8	0
sbic_offensiveness	safety	tasksource/sbic	cc-by-4.0	human	6,590	446	3	0
scruples_who_is_wrong	safety	tasksource/scruples	apache-2.0	votes	25,542	1,344	8	0
spml_prompt_injection	safety	reshabhs/SPML_Chatbot_Prompt_Injection	mit	rules	6,889	363	2	0
told_br_toxicity	safety	mteb/BrazilianToxicTweetsClassification	cc-by-sa-4.0	human	6,551	354	2	0
turkish_hatemap_tone	safety	turkish-nlp-suite/TurkishHateMap	cc-by-sa-4.0	human	34,105	1,795	10	0
tweets_hate_speech	safety	tweets-hate-speech-detection/tweets_hate_speech_detection	gpl-3.0	human	4,057	214	1	0
vihos_offensive_spans	safety	phusroyal/ViHOS	mit	human	8,388	441	3	0
wikipedia_detox_votes	safety	tasksource/wikipedia-detox-votes tasksource/jigsaw_toxicity	cc0-1.0	votes	9,953	526	3	0
banking77_intent	support	legacy-datasets/banking77	cc-by-4.0	human	9,647	508	3	0
bitext_support_intent	support	bitext/Bitext-customer-support-llm-chatbot-training-dataset	cdla-sharing-1.0	rules	21,456	1,129	6	0
bitext_verticals_intent	support	bitext/Bitext-retail-banking-llm-chatbot-training-dataset bitext/Bitext-insurance-llm-chatbot-training-dataset bitext/Bitext-telco-llm-chatbot-training-dataset bitext/Bitext-travel-llm-chatbot-training-dataset bitext/Bitext-mortgage-loans-llm-chatbot-training-dataset bitext/Bitext-wealth-management-llm-chatbot-training-dataset bitext/Bitext-media-llm-chatbot-training-dataset bitext/Bitext-hospitality-llm-chatbot-training-dataset bitext/Bitext-events-ticketing-llm-chatbot-training-dataset	cdla-sharing-1.0	rules	101,506	5,342	30	0
clinc_intent	support	clinc/clinc_oos	cc-by-3.0	human	12,923	682	4	0
crosswoz_dialogue_acts	support	GEM/CrossWOZ	apache-2.0	human	34,420	1,812	10	0
it_support_queue	support	tasksource/it-support-tickets	cc-by-4.0	human	883	46	0	0
massive_intent	support	mteb/amazon_massive_intent mteb/amazon_massive_scenario	cc-by-4.0	human	149,852	7,893	45	0
minds14_banking_intent	support	PolyAI/minds14	cc-by-4.0	human	5,934	313	2	0
mtop_intent	support	tasksource/mtop	cc-by-sa-4.0	human	67,976	3,592	21	0
multiwoz_turn_acts	support	ConvLab/multiwoz21	apache-2.0	human	41,758	2,198	13	0
risawoz_dialogue_acts	support	GEM/RiSAWOZ	cc-by-4.0	human	38,449	2,024	12	0
sgd_dialogue_state	support	GEM/schema_guided_dialog	cc-by-sa-4.0	human	149,991	7,895	45	0
support_tickets_sim	support	s2pidape/support-ticket-dataset	cc-by-4.0	simulator	76,631	4,033	23	0
support_urgency	support	decisionmodelhub/support-urgency	cc-by-4.0	rules	0	0	0	786
taskmaster1_turn_slot	support	ConvLab/tm1	cc-by-4.0	human	64,272	3,383	19	0
taskmaster2_turn_slot	support	ConvLab/tm2	cc-by-4.0	human	73,846	3,887	22	0
taskmaster3_turn_slot	support	ConvLab/tm3	cc-by-4.0	human	60,370	3,178	18	0
acord_clause_relevance	text	tasksource/acord	cc-by-4.0	human	2,856	150	1	0
aims_modern_slavery_criteria	text	mila-ai4h/AIMS.au	cc-by-4.0	human	34,999	1,842	11	0
allegro_review_stars	text	legacy-datasets/allegro_reviews	cc-by-sa-4.0	human	8,532	449	3	0
allocine_reviews	text	tblard/allocine	mit	human	28,347	1,492	9	0
amazon_counterfactual	text	mteb/amazon_counterfactual	cc-by-4.0	human	7,289	384	2	0
amazon_review_polarity	text	fancyzhx/amazon_polarity	apache-2.0	human	44,920	2,364	13	0
arabic_sarcasm	text	iabufarha/ar_sarcasm	mit	human	7,355	387	2	0
argkp_key_point_match	text	NLP-Debater-Project/IBM-Debater-ArgKP ibm-research/argument_quality_ranking_30k	mit	human	1,131	60	1	0
args_me_stance	text	webis/args_me	cc-by-4.0	human	35,975	1,893	11	0
argument_quality	text	ibm-research/argument_quality_ranking_30k	cc-by-3.0	votes	9,565	503	3	0
arxiv_primary_category	text	gfissore/arxiv-abstracts-2021	cc0-1.0	rules	67,489	3,553	20	0
aspect_sentiment	text	Multilingual-NLP/M-ABSA	apache-2.0	human	78,745	4,144	24	0
b2w_product_reviews	text	ruanchaves/b2w-reviews01	cc-by-4.0	human	61,030	3,212	18	0
barec_arabic_readability	text	CAMeL-Lab/BAREC-Corpus-v1.0	cc-by-sa-4.0	human	50,009	2,632	15	0
bbc_hindi_news_topic	text	midas/bbc_hindi_nli	mit	rules	3,721	196	1	0
beyazperde_film_review_stars	text	turkish-nlp-suite/beyazperde-all-movie-reviews	cc-by-sa-4.0	human	33,485	1,762	10	0
bias_in_bios_profession	text	LabHC/bias_in_bios	mit	rules	51,071	2,688	15	0
blimp_grammaticality	text	nyu-mll/blimp	cc-by-4.0	rules	8,813	464	3	0
brighter_emotion_intensity	text	brighter-dataset/BRIGHTER-emotion-intensities	cc-by-4.0	human	9,157	482	3	0
brighter_emotions	text	brighter-dataset/BRIGHTER-emotion-categories	cc-by-4.0	human	4,424	256	3	0
bundestag_speech_party	text	threite/Bundestag-v2	cc0-1.0	rules	37,949	1,997	11	0
case_hold	text	coastalcph/lex_glue	cc-by-4.0	rules	23,583	1,241	7	0
cedr_russian_emotions	text	sagteam/cedr_v1	apache-2.0	human	4,900	258	1	0
cefr_sentence_level	text	edesaras/CEFR-Sentence-Level-Annotations	mit	human	0	0	0	1,500
chid_idiom_cloze	text	thu-coai/chid	apache-2.0	rules	63,280	3,331	19	0
claim_stance	text	ibm-research/claim_stance	cc-by-3.0	human	891	47	0	0
clarin_emo_polish_emotions	text	clarin-knext/CLARIN-Emo	cc-by-4.0	human	5,057	264	2	0
clickbait_detection	text	marksverdhei/clickbait_title_classification	mit	heuristic	8,525	449	3	0
coat_machine_text	text	RussianNLP/coat	apache-2.0	rules	29,490	1,552	9	0
coedit_edit_type	text	grammarly/coedit	apache-2.0	rules	64,926	3,417	19	0
comi_lingua_hinglish	text	LingoIITGN/COMI-LINGUA	cc-by-4.0	human	21,067	1,109	6	0
companyweb_sic_division	text	pborchert/CompanyWeb	cc-by-4.0	human	70,352	3,703	21	0
comve_sense_making	text	tasksource/comve	cc-by-sa-4.0	human	10,632	560	3	0
conceptnet_relation	text	conceptnet5/conceptnet5	cc-by-4.0	human	149,996	7,895	45	0
cuad_clause_presence	text	theatticusproject/cuad	cc-by-4.0	human	7,037	370	2	0
cultura_italia_description_quality	text	biglam/cultural_heritage_metadata_accuracy	cc-by-4.0	human	21,274	1,120	6	0
dactyl_machine_generated	text	ShantanuT01/DACTYL	mit	rules	29,823	1,570	9	0
dbpedia_ontology	text	fancyzhx/dbpedia_14	cc-by-sa-3.0	rules	69,952	3,682	21	0
discovery_discourse_marker	text	sileod/discovery	apache-2.0	rules	150,000	7,895	45	0
docred_relations	text	thunlp/docred	mit	human	10,313	543	3	0
engineering_design_fact_relation	text	siddharthl1293/engineering_design_facts	mit	human	38,740	2,039	12	0
fakenewscorpus_source_type	text	andyP/fake_news_en_opensources	apache-2.0	rules	63,114	3,322	19	0
fewnerd_entity_types	text	DFKI-SLT/few-nerd	cc-by-sa-4.0	human	84,733	4,471	26	0
fignews_bias_propaganda	text	CAMeL-Lab/FIGNEWS-2024	cc-by-sa-4.0	human	6,603	348	2	0
financial_tweet_sentiment	text	zeroshot/twitter-financial-news-sentiment	mit	human	9,221	485	3	0
financial_tweet_topic	text	zeroshot/twitter-financial-news-topic	mit	human	0	0	0	1,500
finer_xbrl_tag	text	nlpaueb/finer-139	cc-by-sa-4.0	human	25,343	1,334	8	0
foodex2_base_term	text	disi-unibo-nlp/feast-foodex2	cc-by-4.0	human	18,021	949	5	0
formality	text	osyvokon/pavlick-formality-scores	cc-by-3.0	human	8,205	432	2	0
fungi_character_category	text	Atsushi/fungi_diagnostic_chars_comparison_japanese	cc-by-4.0	heuristic	61,105	3,216	18	0
fungi_character_shared	text	Atsushi/fungi_diagnostic_chars_comparison_japanese	cc-by-4.0	human	33,128	1,880	12	0
geotagged_posts_same_region	text	yachay/text_coordinates_regions	mit	rules	5,829	307	2	0
go_emotions	text	google-research-datasets/go_emotions	apache-2.0	votes	7,744	411	2	0
hc3_human_vs_ai	text	Hello-SimpleAI/HC3 Hello-SimpleAI/HC3-Chinese	cc-by-sa-4.0	rules	20,436	1,076	6	0
hh_it_vacancy_experience	text	MrAlexGov/it-job-title-skills-ru	cc-by-4.0	human	20,867	1,099	6	0
hh_it_vacancy_key_skills	text	MrAlexGov/it-job-title-skills-ru	cc-by-4.0	human	1,422	75	3	0
hiner_hindi_entity_types	text	cfilt/HiNER-original	cc-by-sa-4.0	human	62,036	3,266	19	0
humor_detection	text	CreativeLang/ColBERT_Humor_Detection	cc-by-2.0	heuristic	21,168	1,115	6	0
humset_pillar	text	nlp-thedeep/humset	apache-2.0	human	14,345	755	4	0
humset_sector	text	nlp-thedeep/humset	apache-2.0	human	19,040	1,053	6	0
injury_narrative_event_code	text	mayerantoine/injury-narrative-coding	apache-2.0	human	36,954	1,945	11	0
irc_reply_parent	text	jkkummerfeld/irc_disentangle	cc-by-4.0	human	38,002	2,000	11	0
jglue_jsts	text	github:yahoojapan/JGLUE	cc-by-sa-4.0	human	5,356	295	4	0
kinopoisk_review_grade	text	blinoff/kinopoisk	mit	human	17,389	915	5	0
ledgar_clause_type	text	coastalcph/lex_glue	cc-by-4.0	heuristic	26,820	1,412	8	0
legisubject_br_subject	text	ronunes/LegiSubject-Br-Summaries	mit	human	23,904	1,258	7	0
llmtrace_ai_text	text	iitolstykh/LLMTrace_classification	apache-2.0	rules	36,610	1,929	11	0
loc_classification_class	text	jimfhahn/lc-bib-embeddings	cc0-1.0	human	70,347	3,702	21	0
lok_sabha_answering_ministry	text	opensansad/lok-sabha-qa	cc-by-4.0	rules	41,408	2,179	12	0
lok_sabha_starred_question	text	opensansad/lok-sabha-qa	cc-by-4.0	rules	7,650	464	3	0
maga_machine_generated	text	anyangsong/MAGA anyangsong/MAGA-cn	mit	rules	31,666	1,667	10	0
mega_acceptability_rating	text	tasksource/mega-acceptability-v2	apache-2.0	human	63,012	3,316	19	0
mexican_tweet_emoji	text	guillermoruiz/MexEmojis	mit	rules	83,229	4,381	25	0
mexican_tweet_state	text	guillermoruiz/MexEmojis	mit	rules	935	49	9	0
news_category	text	heegyu/news-category-dataset	cc-by-4.0	rules	48,482	2,564	15	0
newsfane_arabic_entity_types	text	asas-ai/NewsFANE	cc-by-sa-3.0	human	3,311	174	1	0
nlpcc_weibo_stance	text	strombergnlp/nlpcc-stance	cc-by-4.0	human	2,648	139	1	0
nyt_ingredient_span_role	text	napsternxg/nyt_ingredients	apache-2.0	human	33,591	1,768	10	0
ontolama_subsumption	text	krr-oxford/OntoLAMA	apache-2.0	rules	44,922	2,364	13	0
opentender_cpv_category	text	clips/mteb-nl-opentender-cls-pr	cc-by-4.0	human	15,794	831	5	0
patent_phrase_similarity	text	tasksource/patent-phrase-similarity	cc-by-4.0	human	38,135	2,007	11	0
phishing_email	text	zefang-liu/phishing-email-dataset	lgpl-3.0	human	8,347	439	3	0
plod_abbreviation_span	text	surrey-nlp/PLOD-filtered	cc-by-sa-4.0	rules	34,067	1,793	10	0
politeness	text	Cleanlab/stanford-politeness	mit	human	1,733	91	1	0
pwc_paper_task	text	J0nasW/paperswithcode	mit	human	20,478	1,110	6	0
pwc_research_area	text	J0nasW/paperswithcode	mit	human	7,171	377	4	0
raid_machine_generated	text	liamdugan/raid	mit	rules	8,528	450	3	0
readability_grade	text	agentlans/readability	cc0-1.0	heuristic	91,676	4,830	28	0
redfm_relation	text	Babelscape/REDFM	cc-by-sa-4.0	human	21,074	1,109	6	0
rsg_russe_word_sense	text	RussianNLP/russian_super_glue	mit	human	8,979	473	3	0
ru_headline_topic	text	ai-forever/headline-classification	mit	rules	21,094	1,110	6	0
ru_product_review_sentiment	text	ai-forever/ru-reviews-classification	apache-2.0	human	23,210	1,222	7	0
ru_scibench_grnti_field	text	ai-forever/ru-scibench-grnti-classification	mit	human	18,686	983	6	0
rucola_acceptability	text	RussianNLP/rucola	apache-2.0	human	4,317	227	1	0
sec_10k_sentence_section	text	JanosAudran/financial-reports-sec	apache-2.0	rules	55,146	2,903	17	0
sec_8k_item_type	text	ZipLime/sec-8k-events	apache-2.0	rules	30,265	1,593	9	0
semeval_cause_effect	text	ChristianHugD/semeval2010t8	mit	human	2,150	113	1	0
sentfin_entity_sentiment	text	tasksource/sentfin	mit	human	12,512	659	4	0
swiss_court_legal_area	text	voilaj/swiss-caselaw	cc0-1.0	human	13,042	686	4	0
ted_cpv_division	text	Rayenlat/cpv-match-corpus	cc-by-4.0	human	65,890	3,468	20	0
topical_chat_knowledge_source	text	Conversational-Reasoning/Topical-ChatASR	cdla-sharing-1.0	human	112,087	5,900	34	0
topical_chat_turn_sentiment	text	Conversational-Reasoning/Topical-ChatASR	cdla-sharing-1.0	human	18,809	1,194	14	0
turblimp_grammaticality	text	juletxara/turblimp	cc-by-4.0	rules	6,675	351	2	0
turkish_customer_review_stars	text	turkish-nlp-suite/MusteriYorumlari	cc-by-sa-4.0	human	70,087	3,690	21	0
turkish_supplement_review_stars	text	turkish-nlp-suite/vitamins-supplements-reviews	cc-by-sa-4.0	human	82,408	4,337	25	0
twitteremo_polish_emotions	text	clarin-pl/twitteremo	gpl-3.0	human	10,203	537	3	0
uk_election_target_sentiment	text	tasksource/uk-election	cc-by-4.0	human	8,136	428	2	0
ukpconvarg_convincing	text	tasksource/ukpconvarg	cc-by-4.0	human	9,721	512	3	0
unarxive_imrad_section	text	saier/unarXive_imrad_clf	cc-by-sa-4.0	rules	70,339	3,702	21	0
unfair_tos_clauses	text	coastalcph/lex_glue	cc-by-4.0	human	596	31	0	0
us_headline_outlet	text	dnakhla/us-news-headlines-enriched	cc-by-4.0	rules	34,228	1,802	10	0
us_headline_section	text	dnakhla/us-news-headlines-enriched	cc-by-4.0	rules	25,471	1,341	8	0
us_politician_post_party	text	thien/political	mit	rules	44,921	2,364	13	0
usajobs_gs_grade	text	abigailhaddad/usajobs-scraping	cc0-1.0	human	130,779	6,884	39	0
usajobs_occupational_group	text	abigailhaddad/usajobs-scraping	cc0-1.0	human	44,123	2,742	21	0
vabb_discipline	text	clips/mteb-nl-vabb-mlcls-pr	cc-by-4.0	human	4,026	212	1	0
vd_early_print_genre	text	SBB/VD-Metadata	cc0-1.0	human	40,970	2,156	12	0
vietnamese_news_section	text	NamSyntax/vietnamese-news-classification	apache-2.0	rules	70,420	3,706	21	0
vn_legal_document_area	text	tmquan/thuvienphapluat-vn-vbpl	cc-by-4.0	human	62,938	3,313	19	0
vn_legal_document_issuer	text	tmquan/thuvienphapluat-vn-vbpl	cc-by-4.0	human	36,140	2,103	13	0
vn_legal_document_type	text	tmquan/thuvienphapluat-vn-vbpl	cc-by-4.0	human	44,197	2,765	19	0
wiki_deletion_policy	text	copenlu/wiki-stance	cc-by-sa-4.0	human	55,801	2,937	17	0
wiki_deletion_stance	text	copenlu/wiki-stance	cc-by-sa-4.0	human	55,303	3,223	20	0
wiki_paragraphs_same_section	text	dennlinger/wiki-paragraphs	cc-by-sa-3.0	rules	44,922	2,364	13	0
wikimedqa_condition	text	sileod/wikimedqa	apache-2.0	rules	91,359	4,808	27	0
wikisqe_quality_flag	text	ando55/WikiSQE	cc-by-sa-4.0	human	70,223	3,696	21	0
wildberries_review_rating	text	nyuuzyou/wb-feedbacks	cc0-1.0	human	149,866	7,893	45	0
wordnet_noun_hypernym	text	Hierarchy-Transformers/WordNetNoun	apache-2.0	rules	24,452	1,288	7	0
yandex_geo_review_rating	text	d0rj/geo-reviews-dataset-2023	mit	human	44,917	2,364	13	0
zhihu_question_topics	text	zirui3/zhihu_qa	cc-by-4.0	human	14,079	741	4	0

Limitations

• Some tasks use heuristic, rule-based or simulated labels (Labels column).
• Safety and toxicity tasks contain offensive, hateful and sexual text by design.
• Small sources stay small: the mix target never moves a task beyond 2× its base weight.