datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
CMPR_LTS_SOFT_TOME_ADJ_CTDchess-soft-sf19
avewright/chess-soft-sf19
Official Stockfish 19 MultiPV soft targets. This release supersedes the 25k
pilot. It is not a filter of chess-soft-multipv-lichess or
chess-soft-100m-disagreements.
2,010,006 rows. Source id 4. Vocab compact (1968).
Mix (as generated)
origin
rows
note
self-play (origin=1)
0
SF19 vs SF19, ε=0.20, book + 4 random legal
relabel (origin=0)
0
existing local boards, new SF19 labels
frozen eval
10,000
split=1 in… See the full description on the dataset page: https://huggingface.co/datasets/avewright/chess-soft-sf19.United_States_State_Legislation_with_SummariesTest Push
italic-softkd-pool
italic-softkd-pool
The exact training data of idealab-cs2/zagreus-0.4B-italic-softkd: 21,606 Italian multiple-choice questions with committee soft labels. One soft-KD training run from mii-llm/zagreus-0.4B-ita on the train split reaches 0.4787 on the full ITALIC 10K (official harness, 5-shot fast, temperature 0), from a 0.2802 base.
train is the full pool; the other three splits partition it by provenance:
split
rows
contents
train
21,606
the full training file (union… See the full description on the dataset page: https://huggingface.co/datasets/idealab-cs2/italic-softkd-pool.software-strategist-v1
Software Fundamentals — Strategy Knowledge Base
A language-agnostic knowledge base of software engineering fundamentals, paired with a synthetic instruction-tuning dataset (~13,500 examples) for training small language models (SLMs) as software engineering strategists.
The trained model takes a description of a coding situation and routes it to relevant concepts, outputting synthesized strategic guidance as structured JSON.
Dataset Summary
This dataset provides ~13… See the full description on the dataset page: https://huggingface.co/datasets/jtregunna/software-strategist-v1.arxiv-software-repo-links
arXiv Software Repository Links
A dataset mapping arXiv papers (via DOI) to software repositories they reference or which are implementations of the work. Additionaly includes co-citation analysis and community clustering.
Implementation detection is conducted using the evamxb/dev-author-em-clf model from sci-soft-models
Quick Start
from datasets import load_dataset
# Load DOI-to-repo links
links = load_dataset("cometadata/arxiv-software-repo-links", "links")
#… See the full description on the dataset page: https://huggingface.co/datasets/cometadata/arxiv-software-repo-links.k12-standards-instruction-tasks
K-12 Curriculum Tasks (generated)
2,489 generated instruction/input/output records covering five curriculum tasks:
assessment creation, learning objective generation, misconception detection, standard
explanation, and standards Q&A. Content is predominantly mathematics.
Important: the name is misleading
Despite the name, this dataset contains no school directory data. There are four
columns - task, input, output, metadata - and no staff, principal, or school… See the full description on the dataset page: https://huggingface.co/datasets/robworks-software/k12-standards-instruction-tasks.soft-toy-wbcd-khlparxiv-software-repo-links-datacite-enrichment-format
arXiv Software Repository Links - DataCite Enrichment Format
A collection of metadata enrichments, formatted for DataCite's enrichment API, that add links between arXiv papers (via DOI) and the software repositories they reference or are supplemented by.
Quick Start
from datasets import load_dataset
ds = load_dataset("cometadata/arxiv-software-repo-links-datacite-enrichment-format")
Dataset Description
Each record is a DataCite-style enrichment instruction… See the full description on the dataset page: https://huggingface.co/datasets/cometadata/arxiv-software-repo-links-datacite-enrichment-format.mantinc-catalan-drift
Mantinc — Catalan Drift Benchmark
Descripció (ca)
Mantinc és un banc de proves que avalua si un model de llenguatge continua
responent en català quan el missatge, la conversa prèvia o el context recuperat
l'empenyen a fer-ho en una altra llengua, normalment el castellà o l'anglès.
Dataset Description
Mantinc is a benchmark that measures whether a language model keeps answering
in Catalan when the prompt, prior conversation, or retrieved context… See the full description on the dataset page: https://huggingface.co/datasets/softcatala/mantinc-catalan-drift.alt-kotlin-source-1.9kkSoftware-Architectural-FrameworksSoftware-Architectural-Frameworks
I am releasing a small dataset covering topics related to Frameworks under Software-Architecture.
I have included following topics:
TOGAF
Zachman Framework
IEEE 1471
Matrix-based approach to architecture development
Significance of IEEE 1471 (ISO/IEC 42010)
Benefits of employing architectural frameworks
and Many More!
This dataset can be useful in LLM development. Also those who are working on developing Software development related LLMs then this dataset can… See the full description on the dataset page: https://huggingface.co/datasets/ajibawa-2023/Software-Architectural-Frameworks.NicheHazardQANew Paper! 🎉 🎊
🎉🎊 Released new paper on AI safety! Accepted at ACL 2024.🎉 🎊
Check out our new paper Safety Arithmetic at https://arxiv.org/abs/2406.11801v1 👈
We introduce safety arithmetic, a test-time solution to bring safety back to your custom AI models. Recent studies showed LLMs are prone to elicit harm when fine-tuned or edited with new knowledge. Safety arithmetic can be solved by first removing harm direction in parameter space and then steering the latent… See the full description on the dataset page: https://huggingface.co/datasets/SoftMINER-Group/NicheHazardQA.optimot-linguistic-data
Optimot Linguistic Data
This dataset contains 4,011 entries extracted from the public Optimot linguistic consultation service of the Departament de Política Lingüística, Generalitat de Catalunya.
Each record addresses a Catalan language question or linguistic topic and includes an explanation, source metadata, and a direct source URL when available.
Data
The dataset is provided as JSON Lines:
optimot.jsonl
Each row contains:
Fitxa: Optimot card identifier.… See the full description on the dataset page: https://huggingface.co/datasets/softcatala/optimot-linguistic-data.alt-kotlin-source-1.4kkmath-soft-tokens
Math Soft Tokens Dataset
Contains training steps: numinamath15_step_11_fixed.
gbag-bench
GBAG-Bench — Grounded BI Answer Generation
A public benchmark for the step after the SQL: how faithfully an LLM interprets a query result into a natural-language answer.
NL2SQL measures half the problem. GBAG measures the other half.
📂 GitHub (harness, judge, leaderboard): softisight/gbag-bench
📊 Live leaderboard: LEADERBOARD.md
📐 Metric & rubric: METRIC.md
🪪 License: MIT (questions & harness) — bundled SQLite samples retain their original licenses
Why this… See the full description on the dataset page: https://huggingface.co/datasets/softisight-ai/gbag-bench.arxiv-software-repo-links
arXiv Software Repository Links
A dataset mapping arXiv papers (via DOI) to software repositories they reference or which are implementations of the work. Additionaly includes co-citation analysis and community clustering.
Implementation detection is conducted using the evamxb/dev-author-em-clf model from sci-soft-models
Quick Start
from datasets import load_dataset
# Load DOI-to-repo links
links = load_dataset("cometadata/arxiv-software-repo-links", "links")
#… See the full description on the dataset page: https://huggingface.co/datasets/rafidirtiza/arxiv-software-repo-links.agentic-software-conformance
TeaQL Agentic Software Conformance
Machine-readable evidence for the TeaQL Harness: semantic-model evaluation,
generated artifacts, seven language-native runtimes, executable examples, and
cross-language conformance checks.
This is an evidence dataset, not a leaderboard and not a collection of
unverified model claims. Each row identifies its evidence level, exact source,
verification date, revisions where available, command or gate, result, and
important qualifications. The… See the full description on the dataset page: https://huggingface.co/datasets/teaql/agentic-software-conformance.texas-k12-curriculum-standards-teks
Texas K-12 Curriculum Standards (TEKS-derived)
15,040 generated learning-objective records organized around the Texas Essential
Knowledge and Skills (TEKS) taxonomy, spanning core academic subjects, Career & Technical
Education clusters, and specialized program areas.
How this was built (read this first)
These records are programmatically generated, not transcribed from official standards
documents. A generator took a standards taxonomy - codes, grade levels… See the full description on the dataset page: https://huggingface.co/datasets/robworks-software/texas-k12-curriculum-standards-teks.repro-softmax-as-linear-attention-in-the-large-prompt-regime-a-measure-based-perspecti-traces
Agent traces
Agent sessions published from a Trackio Logbook.
nanoset
Sourceworks NanoSet
NanoSet is an experimental dataset where the main goal is to create a usable chatbot through less training data.
What is in NanoSet?
NanoSet is divded into 3 major sections, containg 36 entries divided into 6 sub-topics. The structure creates 108 total lines of training data, which may be subject to change in the future. The following is a visual on the structure:
108 entries total
3 Sections, each with 36 entries:
Chat Basics (Greetings, Jokes, etc.)… See the full description on the dataset page: https://huggingface.co/datasets/srcworks-software/nanoset.file-format-software-version-compatibility
GIS vector format capabilities and limitations in GDAL
Canonical, always-current version: https://referencesource.org/file-format-software-version-compatibility/
Machine-readable: https://referencesource.org/file-format-software-version-compatibility/data.json — this mirror is a point-in-time copy.
Last verified: 2026-08-04
Stale after: 2027-01-31 (past this date, prefer the canonical copy —
it re-verifies on a cadence this snapshot does not)
Records: 3
Key capabilities… See the full description on the dataset page: https://huggingface.co/datasets/referencesource/file-format-software-version-compatibility.software-end-of-support
Software end-of-support dates
Canonical, always-current version: https://referencesource.org/software-end-of-support/
Machine-readable: https://referencesource.org/software-end-of-support/data.json — this mirror is a point-in-time copy.
Last verified: 2026-08-03
Stale after: 2026-11-01 (past this date, prefer the canonical copy —
it re-verifies on a cadence this snapshot does not)
Records: 68
Release, end-of-active-support and end-of-security-support dates per major version of… See the full description on the dataset page: https://huggingface.co/datasets/referencesource/software-end-of-support.software-interface-version-compatibility
Software interface version compatibility
Canonical, always-current version: https://referencesource.org/software-interface-version-compatibility/
Machine-readable: https://referencesource.org/software-interface-version-compatibility/data.json — this mirror is a point-in-time copy.
Last verified: 2026-08-11
Stale after: 2026-10-10 (past this date, prefer the canonical copy —
it re-verifies on a cadence this snapshot does not)
Records: 248
Which versions of common software… See the full description on the dataset page: https://huggingface.co/datasets/referencesource/software-interface-version-compatibility.soft-label-qnlitt640c-seedsweep-soft-final-s29-v1
TT639G Recombined Tiny Assistant v1
Recombines isolated proof rungs:
TT638D code behavior + dyadic/Mercy proof upstream
TT639E2 context-copy behavior
TT639F3 task-routing behavior
simple rule/Q&A behavior
Blocking dense gates:
seen_combined_pass
upstream_regression_pass
mixed_heldout_pass
anti_collision_pass
Do not run dyadic/Mercy compare unless all four gates pass.
jaquad-sft
softjapan/jaquad-sft
データセットの概要
このデータセットは、JaQuAD（Japanese Question Answering Dataset）をSFT（Supervised Fine-Tuning）形式に変換したものです。日本語の質問応答タスクに特化したinstruction tuning用のデータセットです。
データセットの詳細
言語: 日本語
タスク: 質問応答、instruction tuning
形式: SFT（instruction/input/output）
訓練データ: 31,748件
検証データ: 3,939件
合計: 35,687件
データ形式
各サンプルは以下の形式で構成されています：
{
"id": "tr-000-00-000",
"instruction": "次の文脈に基づいて質問に答えてください。可能なら短く正確に答えてください。",
"input":… See the full description on the dataset page: https://huggingface.co/datasets/softjapan/jaquad-sft.Video-R1-soft-filter-v2
Video-R1-soft-filter-v2
147,850 samples from Video-R1-260k filtered using the complete Kelsey soft filter.
Filter Logic
Remove a sample only if 2+ models can answer it text-only under circular evaluation.
Model
Method
MCQ Coverage
Non-MCQ
GPT-5-mini
Single-pass text-only
263K (full)
✅
Qwen2.5-VL-7B
4-perm circular eval
149,586 MCQ
pass@10
Gemini 3.1 Pro
3-perm circular eval
149,583 MCQ
direct eval (111K)
Comparison
Version
Qwen… See the full description on the dataset page: https://huggingface.co/datasets/AgPerry/Video-R1-soft-filter-v2.software-engineering-and-devops