Append 12,564 vision training rows; preserve previous data and evaluation

Append 38,000 vision training rows; preserve previous data and evaluation

Append 2,000 vision training rows; preserve previous data and evaluation

Restore default, full, vision and filtered-full configs

Restore vision and filtered-full configs

Publish the steered default mix

Publish tasksource-jev-typed-decisions Parquet dataset (full)

Add tasksource-jev-typed-decisions dataset card

Append validated Tasksource decisions: cebab/rating, cebab/food, cebab/ambiance, cebab/service, cebab/noise, fabsa/sentiment, tab/identifier

Publish the steered default mix

Publish tasksource-jev-typed-decisions Parquet dataset (full)

Add tasksource-jev-typed-decisions dataset card

Publish the steered default mix

Publish tasksource-jev-typed-decisions Parquet dataset (full)

Add tasksource-jev-typed-decisions dataset card

Append validated Tasksource decisions: antique, wikiqa, nevir, causal-reasoning/transitivity

Append validated Tasksource decisions: causal-news, slurp/slot-type, topv2/slot-type

Append validated Tasksource decisions: maven-ere/causal, biored/relation, matres/temporal

Clarify English question surfaces and remove parked preference task

Organize dataset card, archive release notes and add decision-model tag

Withdraw low-value Spider aggregation view; retain column selection

Append validated Tasksource decisions: maven-arg/role, maven/event, empathetic-dialogues/emotion, olid/offensive, isarcasm/intended, magpie/idiomatic, preco/coreference, redial/preference, sst5, spider/aggregation, spider/column, semcor, streusle/noun, streusle/verb

Append validated Tasksource decisions: fnc1/stance, perspectrum/stance, ukpconvarg/convincingness, discogem/relation, pubmed-rct/role

Append validated Tasksource decisions: quac/answerability, hatexplain/majority, hotpotqa/supporting-fact, multiwoz22/dialogue-act, multiwoz22/categorical-slot, schema-guided/intent, coqa/yesno, qasper/yesno, topv2/intent, slurp/intent, mcscript2

Append validated Tasksource decisions: maud/consideration, maud/bringdown-timing, maud/ordinary-course-consent, maud/ordinary-course-efforts, maud/antitrust-efforts, maud/specific-performance, ddi/relation, patent-classification/abstract, tabfact, infotabs, squad-v2/answerability, cuad/answerability, hallmarks-of-cancer/yesno, semeval2014-absa/restaurants, semeval2014-absa/laptops, record/candidate, multidogo/finance/intent, multidogo/fastfood/intent, multidogo/airline/intent, multidogo/media/intent, multidogo/software/intent, multidogo/insurance/intent, wikisql/aggregation, wikisql/column, wikisql/operator, acord, edos, evidence-inference

Append validated Tasksource decisions: sentfin, wands, scirepeval/search

Add HTML browser decisions from WebLINX, WebSRC and public-train Mind2Web

Rebuild three English decisions and remove evaluation-only CLadder

Resolve scoped vision licenses and flag research/redistribution restrictions

Clean vision prompt artifacts and fill original dataset license evidence

Expand vision selectively with audited region, preference, correspondence and Super-CLEVR families

Document the multimodal vision config

Add multimodal Jev vision config

Document the multimodal vision config

Add multimodal Jev vision config

Document the multimodal vision config

Add multimodal Jev vision config

Document the multimodal vision config

Add multimodal Jev vision config

Document SNLI-VE label-quality exclusion

Remove SNLI-VE from vision training pilot

Document the multimodal vision config

Add multimodal Jev vision config

Publish reusable historical exclusion evidence

Add filtered-full train configuration (1,305,478 retained decisions) (#2)

Append filtered WebInstruct as canonical Jev decisions

Publish the steered default mix

Publish tasksource-jev-typed-decisions Parquet dataset (full)

Add tasksource-jev-typed-decisions dataset card

Publish the steered default mix