# Text Generation Datasets, Page 3: AI Datasets
Source: https://savrn.com/datasets/tasks/text-generation/page/3
Markdown alternate of the page above; the site index is https://savrn.com/llms.txt

---

SAVRN Model Hub · Datasets by Task

# Text Generation Datasets

142 open-weight text generation datasets in the SAVRN Model Hub, with Deesced, Aaraj Bhatar and FineData publishing the most.

142Datasets

114Publishers

8Licenses

## SAVRN's Take

These are the datasets language models are built and tested on. WikiText and C4 are long-standing pretraining and evaluation corpora, FineWeb and FineWeb-Edu are the modern filtered web sets, and GSM8K and IFEval are the benchmarks labs use to report math and instruction-following scores.

The most common license here is ODC-By, an attribution license for databases, followed by MIT and Apache 2.0. For fine-tuning, quality beats quantity: a small, clean, task-specific set usually does more than another slice of web text. For evaluation, test on your own prompts as well as public benchmarks, since public test sets can leak into training data.

SAVRN Research, 2026-09-18

## Most Downloaded

| Dataset | Publisher | License | Monthly downloads |
| --- | --- | --- | --- |
| [wikitext](https://savrn.com/datasets/wikitext) | Salesforce AI Research | cc-by-sa-3.0 | 1.8M |
| [c4](https://savrn.com/datasets/c4) | Ai2 | odc-by | 1.4M |
| [gsm8k](https://savrn.com/datasets/gsm8k) | OpenAI | mit | 1.2M |
| [fineweb-edu](https://savrn.com/datasets/fineweb-edu) | FineData | odc-by | 426.8k |
| [fineweb](https://savrn.com/datasets/fineweb) | FineData | odc-by | 410k |
| [IFEval](https://savrn.com/datasets/ifeval) | Google | apache-2.0 | 369.7k |
| [finephrase](https://savrn.com/datasets/finephrase) | FineData | odc-by | 324.9k |
| [ettin-pretraining-data](https://savrn.com/datasets/ettin-pretraining-data) | Center for Language and Speech Processing @ JHU | mit | 306k |
| [FineWeb-HQ](https://savrn.com/datasets/fineweb-hq) | EPFL Machine Learning and Optimization Laboratory | odc-by | 278k |
| [wikipedia](https://savrn.com/datasets/wikipedia) | Wikimedia | cc-by-sa-3.0 | 259.7k |

## Licenses

| License | Datasets | Commercial use |
| --- | --- | --- |
| mit | 33 | Yes |
| apache-2.0 | 29 | Yes |
| not stated | 21 | Not stated |
| other | 16 | Read the license |
| odc-by | 14 | Read the license |
| cc-by-4.0 | 14 | Yes |
| cc0-1.0 | 3 | Read the license |
| cc-by-nc-4.0 | 3 | Not without separate permission |

## Who Publishes Them

| Publisher | Datasets |
| --- | --- |
| [Deesced](https://savrn.com/model-publishers/ameluce) | 9 |
| [Aaraj Bhatar](https://savrn.com/model-publishers/aarajbhattarai) | 7 |
| [FineData](https://savrn.com/model-publishers/huggingfacefw) | 4 |
| [OpenGuardrails](https://savrn.com/model-publishers/openguardrails) | 3 |
| [Huang Liang Hsun](https://savrn.com/model-publishers/lianghsun) | 3 |
| [Anikait Singh](https://savrn.com/model-publishers/asingh15) | 3 |

## All 142 Datasets, Page 3 of 3

Dataset · Text generation

### [pgsoft1042](https://savrn.com/datasets/pgsoft1042)

[Deesced](https://savrn.com/model-publishers/ameluce)

[[赏金女王]官方平台-【PG 赏金女王】福利奖励-【赏金女王】真实评测](tt.26new-of-queenbounty.com.cn.md) —— 承载源站 tt.26new-of-queenbounty.com.cn [[赏金女王]官方平台-【PG 赏金女王】无限胡-【赏金女王】无限胡](tt.26new-of-queenofbounty.com.cn.md) —— 承载源站 tt.26new-of-queenofbounty.com.cn [[赏金女王]官方平台-【赏金女王】娱乐模式-【赏金女王】注册](tt.26new-pg-queenbounty.com.cn.md) —— 承载源站 tt.26new-pg-queenbounty.com.cn [[赏金女王]官方平台-【赏金女王】无脑赢-【赏金女王】信誉担保站](tt.26new-pg-queenofbounty.com.cn.md) —— 承载源站 tt.26new-pg-queenofbounty.com.cn [[赏金女王]官方平台-【PG 赏金女王】卡点技巧分享-【赏金女王】内部机制解析](tt.26new-queenbounty-pg.com.cn.md) —— 承载源站 tt.26new-queenbounty-pg.com.cn [[赏金女王]官方平台-【PG 赏金女王】体验版-【赏金女王】卡点技巧分享](tt.26new-queenbounty-softpg.com.cn.md) —— 承载源站 tt.26new-queenbounty-softpg.com.cn [[赏金女王]官方平台-【PG…

Publicly accessible mit

[View dataset](https://savrn.com/datasets/pgsoft1042)

Dataset · Text generation

### [pgsoft1043](https://savrn.com/datasets/pgsoft1043)

[Deesced](https://savrn.com/model-publishers/ameluce)

[[赏金女王]官方平台-【PG 赏金女王】轻松翻倍-【PG 赏金女王】最新入口地址](wg.26new-of-queenbounty.com.cn.md) —— 承载源站 wg.26new-of-queenbounty.com.cn [[赏金女王]官方平台-【赏金女王】礼包中心-【赏金女王】官网下载安装](wg.26new-of-queenofbounty.com.cn.md) —— 承载源站 wg.26new-of-queenofbounty.com.cn [[赏金女王]官方平台-【PG 赏金女王】好友助力彩金-【赏金女王】无脑赢](wg.26new-pg-queenbounty.com.cn.md) —— 承载源站 wg.26new-pg-queenbounty.com.cn [[赏金女王]官方平台-【赏金女王】精品游戏-【赏金女王】首存送100%](wg.26new-pg-queenofbounty.com.cn.md) —— 承载源站 wg.26new-pg-queenofbounty.com.cn [[赏金女王]官方平台-【PG 赏金女王】豪华礼包-【赏金女王】满100提现](wg.26new-queenbounty-pg.com.cn.md) —— 承载源站 wg.26new-queenbounty-pg.com.cn [[赏金女王]官方平台-【赏金女王】发布页导航-【PG 赏金女王】注册教程](wg.26new-queenbounty-softpg.com.cn.md) —— 承载源站 wg.26new-queenbounty-softpg.com.cn…

Publicly accessible mit

[View dataset](https://savrn.com/datasets/pgsoft1043)

Dataset · Text generation

### [pgsoft1044](https://savrn.com/datasets/pgsoft1044)

[Deesced](https://savrn.com/model-publishers/ameluce)

[[赏金女王]官方平台-【赏金女王】稳定赢-【赏金女王】翻倍活动](xb.26new-of-queenbounty.com.cn.md) —— 承载源站 xb.26new-of-queenbounty.com.cn [[赏金女王]官方平台-【PG 赏金女王】礼包中心-【PG 赏金女王】ios版本入口](xb.26new-of-queenofbounty.com.cn.md) —— 承载源站 xb.26new-of-queenofbounty.com.cn [[赏金女王]官方平台-【PG 赏金女王】独赢盘-【赏金女王】福利奖励](xb.26new-pg-queenbounty.com.cn.md) —— 承载源站 xb.26new-pg-queenbounty.com.cn [[赏金女王]官方平台-【赏金女王】首充-【赏金女王】安全网站](xb.26new-pg-queenofbounty.com.cn.md) —— 承载源站 xb.26new-pg-queenofbounty.com.cn [[赏金女王]官方平台-【赏金女王】爆奖机制-【PG 赏金女王】首提失败补偿金](xb.26new-queenbounty-pg.com.cn.md) —— 承载源站 xb.26new-queenbounty-pg.com.cn [[赏金女王]官方平台-【PG 赏金女王】狂暴爆分实测-【PG 赏金女王】扶持力度大](xb.26new-queenbounty-softpg.com.cn.md) —— 承载源站 xb.26new-queenbounty-softpg.com.cn…

Publicly accessible mit

[View dataset](https://savrn.com/datasets/pgsoft1044)

Dataset · Text generation

### [pgsoft1045](https://savrn.com/datasets/pgsoft1045)

[Deesced](https://savrn.com/model-publishers/ameluce)

[[赏金女王]官方平台-【赏金女王】手机登录-【赏金女王】进阶技巧](xl.26new-of-queenbounty.com.cn.md) —— 承载源站 xl.26new-of-queenbounty.com.cn [[赏金女王]官方平台-【赏金女王】安卓版-【赏金女王】哪个更容易赢](xl.26new-of-queenofbounty.com.cn.md) —— 承载源站 xl.26new-of-queenofbounty.com.cn [[赏金女王]官方平台-【PG 赏金女王】一把翻倍实测记录-【PG 赏金女王】随便赢](xl.26new-pg-queenbounty.com.cn.md) —— 承载源站 xl.26new-pg-queenbounty.com.cn [[赏金女王]官方平台-【赏金女王】热门游戏-【PG 赏金女王】爽玩](xl.26new-pg-queenofbounty.com.cn.md) —— 承载源站 xl.26new-pg-queenofbounty.com.cn [[赏金女王]官方平台-【赏金女王】高额返点-【赏金女王】实测稳定玩法](xl.26new-queenbounty-pg.com.cn.md) —— 承载源站 xl.26new-queenbounty-pg.com.cn [[赏金女王]官方平台-【PG 赏金女王】手机登录-【PG 赏金女王】入口网站](xl.26new-queenbounty-softpg.com.cn.md) —— 承载源站 xl.26new-queenbounty-softpg.com.cn…

Publicly accessible mit

[View dataset](https://savrn.com/datasets/pgsoft1045)

Dataset · Text generation

### [pgsoft1046](https://savrn.com/datasets/pgsoft1046)

[Deesced](https://savrn.com/model-publishers/ameluce)

[[赏金女王]官方平台-【赏金女王】注册就送-【赏金女王】免费领取试玩](xt.26new-of-queenbounty.com.cn.md) —— 承载源站 xt.26new-of-queenbounty.com.cn [[赏金女王]官方平台-【PG 赏金女王】正规网站-【赏金女王】秒爆](xt.26new-of-queenofbounty.com.cn.md) —— 承载源站 xt.26new-of-queenofbounty.com.cn [[赏金女王]官方平台-【PG 赏金女王】验证码登录-【赏金女王】试玩体验](xt.26new-pg-queenbounty.com.cn.md) —— 承载源站 xt.26new-pg-queenbounty.com.cn [[赏金女王]官方平台-【赏金女王】规则解析-【PG 赏金女王】纸飞机群](xt.26new-pg-queenofbounty.com.cn.md) —— 承载源站 xt.26new-pg-queenofbounty.com.cn [[赏金女王]官方平台-【PG 赏金女王】新人专区-【PG 赏金女王】快速下载安装](xt.26new-queenbounty-pg.com.cn.md) —— 承载源站 xt.26new-queenbounty-pg.com.cn [[赏金女王]官方平台-【赏金女王】免费高爆-【PG 赏金女王】一倍流水提款](xt.26new-queenbounty-softpg.com.cn.md) —— 承载源站 xt.26new-queenbounty-softpg.com.cn…

Publicly accessible mit

[View dataset](https://savrn.com/datasets/pgsoft1046)

Dataset · Text generation

### [pgsoft1047](https://savrn.com/datasets/pgsoft1047)

[Deesced](https://savrn.com/model-publishers/ameluce)

[[赏金女王]官方平台-【赏金女王】安全认证-【赏金女王】首存送200%](yh.26new-of-queenbounty.com.cn.md) —— 承载源站 yh.26new-of-queenbounty.com.cn [[赏金女王]官方平台-【赏金女王】免费版-【PG 赏金女王】秒翻](yh.26new-of-queenofbounty.com.cn.md) —— 承载源站 yh.26new-of-queenofbounty.com.cn [[赏金女王]官方平台-【PG 赏金女王】盈利方法-【赏金女王】福利领取](yh.26new-pg-queenbounty.com.cn.md) —— 承载源站 yh.26new-pg-queenbounty.com.cn [[赏金女王]官方平台-【赏金女王】上头模式-【赏金女王】活动](yh.26new-pg-queenofbounty.com.cn.md) —— 承载源站 yh.26new-pg-queenofbounty.com.cn [[赏金女王]官方平台-【赏金女王】链接在哪-【赏金女王】赢不停](yh.26new-queenbounty-pg.com.cn.md) —— 承载源站 yh.26new-queenbounty-pg.com.cn [[赏金女王]官方平台-【PG 赏金女王】新人奖励-【PG 赏金女王】快速赢](yh.26new-queenbounty-softpg.com.cn.md) —— 承载源站 yh.26new-queenbounty-softpg.com.cn [[赏金女王]官方平台-【PG…

Publicly accessible mit

[View dataset](https://savrn.com/datasets/pgsoft1047)

Dataset · Text generation

### [pgsoft1048](https://savrn.com/datasets/pgsoft1048)

[Deesced](https://savrn.com/model-publishers/ameluce)

[[赏金女王]官方平台-【PG 赏金女王】高胜率玩法-【赏金女王】老哥推荐](yy.26new-of-queenbounty.com.cn.md) —— 承载源站 yy.26new-of-queenbounty.com.cn [[赏金女王]官方平台-【赏金女王】活动福利-【PG 赏金女王】稳赚](yy.26new-of-queenofbounty.com.cn.md) —— 承载源站 yy.26new-of-queenofbounty.com.cn [[赏金女王]官方平台-【赏金女王】官网入口-【PG 赏金女王】安全平台](yy.26new-pg-queenbounty.com.cn.md) —— 承载源站 yy.26new-pg-queenbounty.com.cn [[赏金女王]官方平台-【赏金女王】移动端入口-【赏金女王】首充活动](yy.26new-pg-queenofbounty.com.cn.md) —— 承载源站 yy.26new-pg-queenofbounty.com.cn [[赏金女王]官方平台-【赏金女王】单机版-【赏金女王】爆分技巧分享](yy.26new-queenbounty-pg.com.cn.md) —— 承载源站 yy.26new-queenbounty-pg.com.cn [[赏金女王]官方平台-【PG 赏金女王】上头-【赏金女王】信誉担保联盟成员](yy.26new-queenbounty-softpg.com.cn.md) —— 承载源站 yy.26new-queenbounty-softpg.com.cn…

Publicly accessible mit

[View dataset](https://savrn.com/datasets/pgsoft1048)

Dataset · Text generation

### [purrence-hyperbullet-100b](https://savrn.com/datasets/purrence-hyperbullet-100b)

[Jon Su](https://savrn.com/model-publishers/jonnester)

A packed pretraining corpus with 100,000,000,000 training input positions and 10,000,000 held-out validation positions, prepared for the recurrent Purrence Hyperbullet language model. It uses a frozen 2,048-ID tokenizer (2,047 ordinary IDs plus structural boundary ID 2,047). Upload status: consider this repository complete only when UPLOADCOMPLETE.json is present. That receipt identifies the immutable data revision whose remote file sizes and checksums were verified. This is a custom memory-mapped corpus, not a Parquet dataset for datasets.loaddataset. Download the core training files with the Hugging Face Hub client: For reproducibility, pass the revision from UPLOADCOMPLETE.json after…

Publicly accessible 10M<n<100M

[View dataset](https://savrn.com/datasets/purrence-hyperbullet-100b)

Dataset · Text generation

### [QWE](https://savrn.com/datasets/qwe)

[SHRIJAN KUMAR TIWARI](https://savrn.com/model-publishers/shrijanagain)

Instruction-tuning dataset with chain-of-thought, merged and deduplicated from multiple public reasoning datasets. Every row is JSON with: - instruction - the user/task prompt - thought - reasoning wrapped in tags (always present) - response - the final answer - hasthinking - always true - sourcedataset - provenance of the row in row order. Sources include OpenThoughts3-1.2M, Mixture-of-Thoughts, Infinity-Instruct and Nemotron-Post-Training-Dataset (v1 and v2). Deduplication is exact full-line (md5), applied globally across all sources. - All model-identity references were normalized to "Speed by SKT AI LABS Private Limited". - A small fraction of rows (mostly from one legacy source) carry…

Publicly accessible apache-2.0 100B<n<1T

[View dataset](https://savrn.com/datasets/qwe)

Dataset · Text generation

### [qwen25-math-sft-long-think7k-v1](https://savrn.com/datasets/qwen25-math-sft-long-think7k-v1)

[Hoang Hai Dang](https://savrn.com/model-publishers/danghoang2005)

2,500 train and 250 validation rows. Long examples have 4,096–7,168 thinking tokens; complete ChatML samples including final answer and EOS fit in 8,192 tokens. Long source: open-r1/OpenR1-Math-220k at e4e141ec9dea9f8326f4d347be56105859b2bd68 (only math-verified generations). Short replay source: danghoang2005/qwen25-math-sft-long-15k-v1 at 32f0afc54a09c4963ae2478cf93871025abda901. Tokenizer: Qwen/Qwen2.5-Math-1.5B at 4a83ca6e4526a4f2da3aa259ec36c259f66b2ab2. Deduplicated across splits and screened against GSM8K, SVAMP, MATH-500, and AIME 2026. Review the source datasets' licenses and terms before reuse or redistribution. See manifest.json for counts and reproducibility settings.

Publicly accessible

[View dataset](https://savrn.com/datasets/qwen25-math-sft-long-think7k-v1)

Dataset · Text generation

### [Qwen3.5-0.8B-Japanese-SFT-v2.5-Data](https://savrn.com/datasets/qwen3-5-0-8b-japanese-sft-v2-5-data)

[Atemoya](https://savrn.com/model-publishers/takenoko12345678)

Takenoko12345678/Qwen3.5-0.8B-Japanese-SFT-v2.5 の追加学習に使ったデータを、学習に使ったそのままの形で公開しています。中身は変えず、1件ごとに出典（source）・ライセンス（license）・文を書いたモデル（generator）の欄だけを足しています。 学習・データ作りのスクリプトは Pitcher プロジェクト の sft/q08bv25/ にあります。 2段目の chatthink 以外の 3,000件は、1段目のデータからの復習用（半分は理系）です。 思考ありの手本は、学習時に「 考える部分 答え」の形にしています（元の Qwen3.5 のチャットテンプレートで enablethinking=True）。 - このデータのために新しく書かせた文は、すべて Qwen/Qwen3.8-27B（Apache 2.0）によるものです。理系の問題の日本語訳と解き方、会話の手本、コードの回答、思考の部分がこれに当たります。tengentoppa・formatdrill は、公開済みの Japanese-SFT-Tengentoppa-Curated-12K の行をそのまま使っています（その元データの作り方は、元データの説明を見てください） - コードの手本の選別には Gemma 4 31B の採点を使っています（文は書いていません）。詳しくは Japanese-Code-Review-OSS の説明を見てください…

Publicly accessible other 10K<n<100K

[View dataset](https://savrn.com/datasets/qwen3-5-0-8b-japanese-sft-v2-5-data)

Dataset · Text generation

### [resume-parser-dataset](https://savrn.com/datasets/resume-parser-dataset)

[Muhammad Zain](https://savrn.com/model-publishers/md-zain)

This repository documents and packages the data pipeline for the resume-parser-model project. The project explores fine-tuning compact language models to turn resume text into structured JSON, using only facts supported by each resume. The folders preserve the stages used to build the dataset: Resume files and their derived text/annotations use matching category folders and resume IDs so the processing trail can be followed across stages. Each JSONL record contains four top-level fields: - resumeid: the stable resume identifier shared across processing stages. - category: the resume's occupation/category label. - resumetext: the extracted text used as model input (X). - targetjson: the…

Publicly accessible

[View dataset](https://savrn.com/datasets/resume-parser-dataset)

Dataset · Text generation

### [smol-software-harbor](https://savrn.com/datasets/smol-software-harbor)

[Edoardo Federici](https://savrn.com/model-publishers/efederici)

1,002 small, verifiable software-engineering tasks in the Harbor format, for training and evaluating small coding agents. templated. About a fifth are search tasks: find the right place in a larger repository with distractor files, then act. Task types (26): fixbug, addfeature, locatefix, edgecase, implementstub, searchanswer, addvalidation, findandchange, addtest, readanswer, fixshellscript, addcliflag, fixsqlquery, renamesymbol, handlemissingfile, tracebackfix, configchange, updatecallsites, writetransformscript, fixbuild, refactorextract, addmaketarget, fixregex, removedeadcode, bumpversion, and more. Splits are made by codebase, never by task. Task names. task.toml names are…

Publicly accessible apache-2.0 1K<n<10K

[View dataset](https://savrn.com/datasets/smol-software-harbor)

Dataset · Text generation

### [smolified-aiexpense](https://savrn.com/datasets/smolified-aiexpense)

[Smolify AI](https://savrn.com/model-publishers/smolify)

This is a synthetic training corpus generated by the Smolify Foundry. It was used to train the corresponding model smolify/smolified-aiexpense. This dataset is a sovereign asset owned by smolify. Generated via Smolify.ai.

Publicly accessible apache-2.0 1K<n<10K

[View dataset](https://savrn.com/datasets/smolified-aiexpense)

Dataset · Text generation

### [stack-v3-docs](https://savrn.com/datasets/stack-v3-docs)

[Takamichi Ohara](https://savrn.com/model-publishers/tokinasin)

Exact duplicate samples were removed after extraction. These subsets are derived from The Stack v3 train dataset, which is released under the ODC-By 1.0 license. Individual source files may have different original licenses; their use must comply with those terms, including attribution where applicable. Repository provenance and per-file license metadata (repopath, commitid, licensetype, and detectedlicenses) are preserved in each sample. License detection may be incomplete or inaccurate.

Publicly accessible odc-by

[View dataset](https://savrn.com/datasets/stack-v3-docs)

Dataset · Text generation

### [SWE-Misbehavior](https://savrn.com/datasets/swe-misbehavior)

[Nguyen Duc Anh](https://savrn.com/model-publishers/duc-anh-2002)

SWE-Misbehavior is a benchmark for measuring boundary violations made by software coding agents, meaning actions that go beyond what the user asked for. Examples include rewriting code around the target function, changing signatures, touching unrelated files, deleting the wrong files, running unrequested commands, or skipping files the agent needed to read. The benchmark has 7,427 task instances across 3 task families and 9 subsets. All of them are built on real-world Python repositories from DevEval (10 domains: Communications, Database, Internet, Multimedia, Scientific-Engineering, Security, Software-Development, System, Text-Processing, Utilities). Each instance comes with 50 paraphrased…

Publicly accessible 1K<n<10K

[View dataset](https://savrn.com/datasets/swe-misbehavior)

Dataset · Text generation

### [syed-med-medical-qa-synthetic-4k](https://savrn.com/datasets/syed-med-medical-qa-synthetic-4k)

[Syed Masood Haji Ahmedyaseen](https://savrn.com/model-publishers/zaiath)

Locally generated synthetic medical question/answer rows, produced with a local DeepSeek-R1 model (via Ollama, $0, no API calls) and used — together with 7 public datasets — to fine-tune 3,002 + 801 + 401 = 4,204 rows — the synthetic share of the 196,846-row SFT set. Every row is one JSON object using one template — the same one the model was trained on (there is no separate system/schema to guess): Whitespace is normalised. Rows carry no end-of-turn token on purpose: the model was trained to stop at the next ### marker and the demo app enforces that too. 1. a topic matrix (condition × country × age group) drove a local DeepSeek-R1 model; 2. every row was validated (template match…

Publicly accessible other 1K<n<10K

[View dataset](https://savrn.com/datasets/syed-med-medical-qa-synthetic-4k)

Dataset · Text generation

### [tb21-dsv4-flash-0731-dsh](https://savrn.com/datasets/tb21-dsv4-flash-0731-dsh)

[OpenGuardrails](https://savrn.com/model-publishers/openguardrails)

Every trial of this one line, in one place: the 89-task main run, both re-run passes, and the scoring scripts. The trajectories are raw and unedited — each step's reasoning, each tool call, and the verifier's own stdout. This is a re-packaging, not a new measurement. The same files were published before, split across two releases, which made the line look incomplete in both: the first release carried the main run without the later re-runs, the second carried 11 re-runs without the main run. Nothing here is new data and no number changes. Both source releases remain up and unmodified. Three tiers, applied by one script (scripts/score4.py) identically to this and the three other lines it was…

Publicly accessible mit n<1K

[View dataset](https://savrn.com/datasets/tb21-dsv4-flash-0731-dsh)

Dataset · Text generation

### [tb21-dsv41-flash-dsh](https://savrn.com/datasets/tb21-dsv41-flash-dsh)

[OpenGuardrails](https://savrn.com/model-publishers/openguardrails)

Raw, unedited agent trajectories for the DeepSeek-V4.1-Flash run over all 89 Terminal-Bench 2.1 tasks, plus every remediation re-run for it and for the two older lines, plus the scoring and analysis scripts. Every number in the write-up can be traced to a file here. Exact McNemar, paired over the same 89 tasks: V4.1 is the only line above 70%, and it is 9 points above V4 on the same scaffold — but 89 tasks cannot carry that to significance. The direction is clear; "proven stronger" is not a claim this sample supports. Applied identically to every line, by one script (scripts/score4.py): 1. as-shipped — what the run actually scored. This is what you get if you deploy the same stack today. 2.…

Publicly accessible mit n<1K

[View dataset](https://savrn.com/datasets/tb21-dsv41-flash-dsh)

Dataset · Text generation

### [tinyDTVi-dataset](https://savrn.com/datasets/tinydtvi-dataset)

[Hoàng Duy](https://savrn.com/model-publishers/hoangduy02071997)

This is the official pretraining dataset for the TinyDTVi-500M causal language model. The corpus is aggressively filtered and deduplicated to maximize the representation of standard Vietnamese text for training resource-constrained models. The initial raw dataset (~135GB) was curated by mixing three high-quality open-source Vietnamese corpora. The approximate mixture ratio is: - 50% CulturaX-Vi (Cleaned web crawl) - 30% VNews (Vietnamese News Corpus) - 20% Vietnamese Wikipedia (Encyclopedia) After strict preprocessing and deduplication, the final dataset size is approximately 85GB (yielding roughly 18 billion unique tokens). To compress the vocabulary down to 50k tokens and focus strictly…

Publicly accessible mit 10B<n<100B

[View dataset](https://savrn.com/datasets/tinydtvi-dataset)

Dataset · Text generation

### [tool-calling-conversations-mu3c3k00](https://savrn.com/datasets/tool-calling-conversations-mu3c3k00)

[Dark Panda](https://savrn.com/model-publishers/dakr-pandas)

An Arena-style dataset of anonymized, multi-turn conversations focused on real-world tool use. It is intended for research, evaluation, and training of models that decide when and how to call tools. Data is organized into date-partitioned Parquet files. Each row contains one complete conversation trace with its tool calls and results. Flexible nested payloads are stored losslessly in metadatajson, inputjson, and outputjson; parse these JSON strings before training or analysis. processing.py provides streaming normalize and decode commands plus importable normalizetrace and decodetrace helpers for arbitrary JSON-compatible contexts. The repository is publicly discoverable, but files require…

Access requested at publisher

[View dataset](https://savrn.com/datasets/tool-calling-conversations-mu3c3k00)

Dataset · Text generation

### [wildchecklists](https://savrn.com/datasets/wildchecklists)

[Yike Wang](https://savrn.com/model-publishers/yikeee)

WildChat-derived checklist/requirements data, excluding examples already present in yikeee/wildchecklistsopenrubricsrubrics. Rows from combinedwildchatrequirements whose instruction appears in yikeee/wildchecklistsopenrubricsrubrics (matched via messages[0].content) were removed. - Removed (instruction overlap, including duplicate instructions): 62,617 - instruction: user prompt - rawresponse: model/raw checklist generation output - requirements: parsed requirements list - origrow: original WildChat row metadata - origrowidx: index in source file - sourcefile: source file path

Publicly accessible apache-2.0

[View dataset](https://savrn.com/datasets/wildchecklists)

## Other Tasks

- [Robotics](https://savrn.com/datasets/tasks/robotics) 123
- [Text Classification](https://savrn.com/datasets/tasks/text-classification) 52
- [Other](https://savrn.com/datasets/tasks/other) 51
- [Text Retrieval](https://savrn.com/datasets/tasks/text-retrieval) 44
- [Question Answering](https://savrn.com/datasets/tasks/question-answering) 40
- [Speech Recognition](https://savrn.com/datasets/tasks/speech-recognition) 35
- [Image to Text](https://savrn.com/datasets/tasks/image-to-text) 30
- [Time Series Forecasting](https://savrn.com/datasets/tasks/time-series-forecasting) 23
- [Tabular Classification](https://savrn.com/datasets/tasks/tabular-classification) 22
- [Image Classification](https://savrn.com/datasets/tasks/image-classification) 18
- [Visual Question Answering](https://savrn.com/datasets/tasks/visual-question-answering) 17
- [Object Detection](https://savrn.com/datasets/tasks/object-detection) 16

[See all](https://savrn.com/datasets/tasks)