datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Zephyrus
ZephyrusBench
ZephyrusBench is a weather-science benchmark released with the paper Zephyrus: An Agentic Framework for Weather Science. It contains 2,230 question-answer pairs across 49 tasks spanning geospatial reasoning, temporal reasoning, forecasting, simulation, climatology, and scientific question answering.Accepted at the International Conference on Learning Representations, 2026.
Paper and Resources
Paper: arXiv
Poster: ICLR 2026 Poster
Code: Rose-STL-Lab/Zephyrus… See the full description on the dataset page: https://huggingface.co/datasets/Rose-STL-Lab/Zephyrus.rosettacode-parsed
Data Origins
Original dataset: https://huggingface.co/datasets/jondurbin/rosettacode-raw/
Cleaner code: https://github.com/the-crypt-keeper/rosettacode-parser
Data Fields
Field
Type
Description
title
string
problem title
task
string
problem description
language
string
solution language/variant
soulution
string
solution source code
Languages
One .jsonl is provided per language group, the sublanguage field in the data denotes the… See the full description on the dataset page: https://huggingface.co/datasets/mike-ravkine/rosettacode-parsed.bridgeTLDR: This dataset is a real-world math tutoring dataset from the NAACL 2024 paper ``Bridging the Novice-Expert Gap via Models of Decision-Making: A Case Study on Remediating Math Mistakes''.
The dataset targets scenarios where the student makes a math mistake.
c_h is the conversation history
c_r is the original tutor's response
c_r_ is the experienced teacher's response
Optionally, there is other interesting metadata from our Bridge method:
e is the student error type that the experienced… See the full description on the dataset page: https://huggingface.co/datasets/rose-e-wang/bridge.Rosebleu-1on1-Dialogues-RP
Rosebleu-1on1-Dialogues-RP
2025/05/17 3人での対話のデータを追加&無駄な改行の削除
@matsuxrさんが公開しているRosebleuデータセットを加工したAratako/Rosebleu-1on1-Dialoguesを元に、キャラクターや作品の設定などを付け加えたうえで、ロールプレイ的な文脈になるように加工したデータセットです。
LLMのファインチューニングにおけるロールプレイングタスクの学習を想定しています。
OpenAI APIのようにroleとcontentのペアの形式となっており、tokenizer.apply_chat_template()によって簡単に各モデルのチャットテンプレートのデータセットへと変換可能です。
データセットの詳細
各キャラの設定や各作品の世界観・あらすじなどをWikipediaやニコニコ大百科からまとめ、ロールプレイ向けにシステムメッセージへと埋め込んでいます。
現在、以下の2パターンのデータセットを用意してあります。主に地の文の処理方法が異なります。… See the full description on the dataset page: https://huggingface.co/datasets/Aratako/Rosebleu-1on1-Dialogues-RP.polaris_rose_rollouts_olmo3-7b_from_qwen3-30b-a3b_cutoff4096_240steps
Cross-tokenizer ROSE rollouts — Olmo-3-7B-Think-SFT ← Qwen3-30B-A3B-Thinking-2507
Every assembled row of a complete 240-step online-ROSE run: 61,440 rows, the teacher's
actual continuation for each, and the token accounting behind it.
The student writes a 4096-token prefix in its own vocabulary (100278). That prefix is
decoded to text, the teacher is shown it under its own chat template, and the teacher's
reply comes back as text and is tokenised into the student's vocabulary.… See the full description on the dataset page: https://huggingface.co/datasets/SeanWang0027/polaris_rose_rollouts_olmo3-7b_from_qwen3-30b-a3b_cutoff4096_240steps.rose_code_samples
rose_code samples (pass@8 rollouts)
vLLM pass@8 samples on the CL-From-Nothing/rose_code train split (23,688 codeforces stdin/stdout
problems), scored by the deepcoder verifier (reward=1.0 iff all test cases pass).
Qwen3-1.7B/ — student model rollouts. 23,688 questions × 8 samples = 189,504 lines.
Qwen3-4B-Thinking-2507/ — teacher model rollouts.
Sampling: temperature 0.7, top_p 0.9, max_tokens 16384, 8 samples/question (pass@8).
Each cluster file holds a contiguous… See the full description on the dataset page: https://huggingface.co/datasets/CL-From-Nothing/rose_code_samples.zero-shot-teacher-feedbackTLDR: Classification + text generation feedback on classroom transcripts.
Is ChatGPT a Good Teacher Coach? Measuring Zero-Shot Performance For Scoring and Providing Actionable Insights on Classroom Instruction
Paper •
Project Page •
Code
Authors: Rose E. Wang and Dorottya Demszky
In the Proceedings of Innovative Use of NLP for Building Educational Applications 2023
Selected as the Ambassador Paper for BEA 2023! 🎉 To be presented at AIED 2024.
If you find… See the full description on the dataset page: https://huggingface.co/datasets/rose-e-wang/zero-shot-teacher-feedback.Turngen
Turngen
A large scale turnwise conversational dataset, with a focus on roleplay data
Data breakdown
Datasets included:
PygmalionAI/PIPPA license: apache-2.0
873380 samples from HuggingfaceTB/smoltalk see repo for licensing information; different subsets have different licenses
2,380 samples from everyday-conversations
50,000 samples from metamathqa-50k
112,000 samples from numina-cot-100k
50,000 samples from self-oss-instruct
36,000 samples from smol-constraints
431,000… See the full description on the dataset page: https://huggingface.co/datasets/rosethelocalfem/Turngen.backtracing
Backtracing: Retrieving the Cause of the Query
Paper •
Code •
Video •
Citation
Authors: Rose E. Wang, Pawan Wirawarn, Omar Khattab, Noah Goodman, Dorottya Demszky
Findings of EACL, Long Paper, 2024.
What is Backtracing?
Many online content portals allow users to ask questions to supplement their understanding (e.g., of lectures or news articles).
While information retrieval (IR) systems may provide answers for such user queries, they do not… See the full description on the dataset page: https://huggingface.co/datasets/rose-e-wang/backtracing.claude-code-rev-benchmark
Benchmark Suite / 基准测试套件
Professional benchmark harness for this restored Claude Code repository.
面向本还原版 Claude Code 仓库的专业级 Benchmark 框架。
Original upstream repository / 原版仓库地址：https://github.com/anthropics/claude-code
Restored source repository / 还原版源码仓库地址：https://github.com/oboard/claude-code-rev
Commands / 命令
bun run bench:smoke - quick hybrid pass for local validation快速混合模式验证（本地冒烟）
bun run bench:full - full hybrid suite完整混合模式基准
bun run bench:offline -… See the full description on the dataset page: https://huggingface.co/datasets/roseking/claude-code-rev-benchmark.rosettabench-150-stratified-compressed
About Dataset
Why?
Used for RosettaBench (A contamination-free benchmark for measuring learning, not memorization).
Preparation
Source: LiveCodeBench (release_v5, AtCoder platform only), accessed via sam-paech/livecodebench-code_generation_lite on Hugging Face.
AtCoder problems were selected for platform consistency and STDIN/STDOUT format compatibility. Problems with fewer than 3 test cases were excluded, leaving a pool of 342 problems.
Sampling: 150… See the full description on the dataset page: https://huggingface.co/datasets/namanbnsl/rosettabench-150-stratified-compressed.new_chymical_wedding_of_christian_rosenkreutz
The New Chymical Wedding of Christian Rosenkreutz Multiturn Conversation Dataset
Overview
This dataset consists of structured multiturn conversations modeled around the esoteric and philosophical themes of "The New Chymical Wedding of Christian Rosenkreutz." The text serves as the foundation for generating dialogues that involve rigorous inquiry into the occult and philosophical.
Objective
The primary objective of this dataset is to facilitate the development… See the full description on the dataset page: https://huggingface.co/datasets/alexandreteles/new_chymical_wedding_of_christian_rosenkreutz.rosetta-ko-math-synth-sft
rosetta-ko-math-synth-sft
Korean-native mathematics data — problems with step-by-step Korean solutions, answer-verified by Math-Verify (LaTeX \boxed{} + symbolic equivalence).
Synthetic data generated with the Qwen3.6-27B teacher model — part of the Rosetta-KO suite for the Rosetta Korean LLM (PoSTMEDIA).
Math Suite
Sibling datasets from the same pipeline (each a separate repo):
repo
format
rosetta-ko-math-synth-sft (this repo)
supervised fine-tuning… See the full description on the dataset page: https://huggingface.co/datasets/PoSTMEDIA/rosetta-ko-math-synth-sft.chymical_wedding_of_christian_rosenkreutz_multiturn
The Chymical Wedding of Christian Rosenkreutz Multiturn Conversation Dataset
Overview
This dataset consists of structured multiturn conversations modeled around the esoteric and philosophical themes of "The Chymical Wedding of Christian Rosenkreutz." The text, known for its deep allegorical content, serves as the foundation for generating dialogues that involve rigorous inquiry into the occult and philosophical.
Objective
The primary objective of this dataset… See the full description on the dataset page: https://huggingface.co/datasets/alexandreteles/chymical_wedding_of_christian_rosenkreutz_multiturn.rosetta-ko-math-synth-sft-think
rosetta-ko-math-synth-sft-think
Korean-native mathematics data — problems with step-by-step Korean solutions, answer-verified by Math-Verify (LaTeX \boxed{} + symbolic equivalence).
Synthetic data generated with the Qwen3.6-27B teacher model — part of the Rosetta-KO suite for the Rosetta Korean LLM (PoSTMEDIA).
Math Suite
Sibling datasets from the same pipeline (each a separate repo):
repo
format
rosetta-ko-math-synth-sft
supervised fine-tuning… See the full description on the dataset page: https://huggingface.co/datasets/PoSTMEDIA/rosetta-ko-math-synth-sft-think.rosetta-ko-chat-synth-sft-think
rosetta-ko-chat-synth-sft-think
Korean-native general-purpose conversational data — diverse everyday instructions with natural Korean answers.
Synthetic data generated with the Qwen3.6-27B teacher model — part of the Rosetta-KO suite for the Rosetta Korean LLM (PoSTMEDIA).
Chat Suite
Sibling datasets from the same pipeline (each a separate repo):
repo
format
rosetta-ko-chat-synth-sft
supervised fine-tuning (user/assistant messages)… See the full description on the dataset page: https://huggingface.co/datasets/PoSTMEDIA/rosetta-ko-chat-synth-sft-think.rosetta-ko-chat-synth-sft
rosetta-ko-chat-synth-sft
Korean-native general-purpose conversational data — diverse everyday instructions with natural Korean answers.
Synthetic data generated with the Qwen3.6-27B teacher model — part of the Rosetta-KO suite for the Rosetta Korean LLM (PoSTMEDIA).
Chat Suite
Sibling datasets from the same pipeline (each a separate repo):
repo
format
rosetta-ko-chat-synth-sft (this repo)
supervised fine-tuning (user/assistant messages)… See the full description on the dataset page: https://huggingface.co/datasets/PoSTMEDIA/rosetta-ko-chat-synth-sft.code_rose_initial_1_7B_SFT_10K_rollouts_Qwen3-4B-Thinking-2507_k12_t0.7_maxtok12288
code_rose_initial_1_7B_SFT_10K — rollouts (Qwen3-4B-Thinking-2507, k=12)
Pass@k completions generated with vLLM over the prefixes in
CL-From-Nothing/code_rose_initial_1_7B_SFT_10K.
Generation config
Model
Qwen3-4B-Thinking-2507
Samples per question (k)
12
Temperature
0.7
top_p
0.9
max_tokens
12288
max_model_len
32768
Questions
7250 (index 0–7249, full split)
Total rows
87000 (7250 × 12)
Generated by complete_prefix_vllm.py… See the full description on the dataset page: https://huggingface.co/datasets/CL-From-Nothing/code_rose_initial_1_7B_SFT_10K_rollouts_Qwen3-4B-Thinking-2507_k12_t0.7_maxtok12288.rosetta-ko-chat-synth-rlvr
rosetta-ko-chat-synth-rlvr
Korean-native general-purpose conversational data — diverse everyday instructions with natural Korean answers.
Synthetic data generated with the Qwen3.6-27B teacher model — part of the Rosetta-KO suite for the Rosetta Korean LLM (PoSTMEDIA).
Chat Suite
Sibling datasets from the same pipeline (each a separate repo):
repo
format
rosetta-ko-chat-synth-sft
supervised fine-tuning (user/assistant messages)… See the full description on the dataset page: https://huggingface.co/datasets/PoSTMEDIA/rosetta-ko-chat-synth-rlvr.rosetta-ko-heritage-synth-cpt
rosetta-ko-heritage-synth-cpt
Korean cultural-heritage data grounded in public heritage records — a cleaned CPT corpus plus source-grounded QA/preference/RLVR sets.
Synthetic data generated with the Qwen3.6-27B teacher model — part of the Rosetta-KO suite for the Rosetta Korean LLM (PoSTMEDIA).
Cultural Heritage Suite
Sibling datasets from the same pipeline (each a separate repo):
repo
format
rosetta-ko-heritage-synth-cpt (this repo)… See the full description on the dataset page: https://huggingface.co/datasets/PoSTMEDIA/rosetta-ko-heritage-synth-cpt.rosetta-ko-law-synth-sft
rosetta-ko-law-synth-sft
Korean legal-domain data grounded in national statutes — a cleaned CPT corpus plus source-grounded QA/preference/RLVR sets.
Synthetic data generated with the Qwen3.6-27B teacher model — part of the Rosetta-KO suite for the Rosetta Korean LLM (PoSTMEDIA).
Law Suite
Sibling datasets from the same pipeline (each a separate repo):
repo
format
rosetta-ko-law-synth-cpt
continued-pretraining corpus (plain text)… See the full description on the dataset page: https://huggingface.co/datasets/PoSTMEDIA/rosetta-ko-law-synth-sft.rosetta-ko-math-synth-dpo
rosetta-ko-math-synth-dpo
Korean-native mathematics data — problems with step-by-step Korean solutions, answer-verified by Math-Verify (LaTeX \boxed{} + symbolic equivalence).
Synthetic data generated with the Qwen3.6-27B teacher model — part of the Rosetta-KO suite for the Rosetta Korean LLM (PoSTMEDIA).
Math Suite
Sibling datasets from the same pipeline (each a separate repo):
repo
format
rosetta-ko-math-synth-sft
supervised fine-tuning… See the full description on the dataset page: https://huggingface.co/datasets/PoSTMEDIA/rosetta-ko-math-synth-dpo.rosetta-ko-math-synth-rlvr
rosetta-ko-math-synth-rlvr
Korean-native mathematics data — problems with step-by-step Korean solutions, answer-verified by Math-Verify (LaTeX \boxed{} + symbolic equivalence).
Synthetic data generated with the Qwen3.6-27B teacher model — part of the Rosetta-KO suite for the Rosetta Korean LLM (PoSTMEDIA).
Math Suite
Sibling datasets from the same pipeline (each a separate repo):
repo
format
rosetta-ko-math-synth-sft
supervised fine-tuning… See the full description on the dataset page: https://huggingface.co/datasets/PoSTMEDIA/rosetta-ko-math-synth-rlvr.rosetta-ko-instruction-following-synth-rlvr
rosetta-ko-instruction-following-synth-rlvr
Korean-native instruction-following data — instructions with verifiable constraints (length, format, keywords, JSON, ...) checked by programmatic verifiers.
Synthetic data generated with the Qwen3.6-27B teacher model — part of the Rosetta-KO suite for the Rosetta Korean LLM (PoSTMEDIA).
Instruction-Following Suite
Sibling datasets from the same pipeline (each a separate repo):
repo
format… See the full description on the dataset page: https://huggingface.co/datasets/PoSTMEDIA/rosetta-ko-instruction-following-synth-rlvr.rosetta-ko-instruction-following-synth-sft-think
rosetta-ko-instruction-following-synth-sft-think
Korean-native instruction-following data — instructions with verifiable constraints (length, format, keywords, JSON, ...) checked by programmatic verifiers.
Synthetic data generated with the Qwen3.6-27B teacher model — part of the Rosetta-KO suite for the Rosetta Korean LLM (PoSTMEDIA).
Instruction-Following Suite
Sibling datasets from the same pipeline (each a separate repo):
repo
format… See the full description on the dataset page: https://huggingface.co/datasets/PoSTMEDIA/rosetta-ko-instruction-following-synth-sft-think.rosetta-ko-law-synth-cpt
rosetta-ko-law-synth-cpt
Korean legal-domain data grounded in national statutes — a cleaned CPT corpus plus source-grounded QA/preference/RLVR sets.
Synthetic data generated with the Qwen3.6-27B teacher model — part of the Rosetta-KO suite for the Rosetta Korean LLM (PoSTMEDIA).
Law Suite
Sibling datasets from the same pipeline (each a separate repo):
repo
format
rosetta-ko-law-synth-cpt (this repo)
continued-pretraining corpus (plain text)… See the full description on the dataset page: https://huggingface.co/datasets/PoSTMEDIA/rosetta-ko-law-synth-cpt.rosetta-ko-law-synth-rlvr
rosetta-ko-law-synth-rlvr
Korean legal-domain data grounded in national statutes — a cleaned CPT corpus plus source-grounded QA/preference/RLVR sets.
Synthetic data generated with the Qwen3.6-27B teacher model — part of the Rosetta-KO suite for the Rosetta Korean LLM (PoSTMEDIA).
Law Suite
Sibling datasets from the same pipeline (each a separate repo):
repo
format
rosetta-ko-law-synth-cpt
continued-pretraining corpus (plain text)… See the full description on the dataset page: https://huggingface.co/datasets/PoSTMEDIA/rosetta-ko-law-synth-rlvr.rosetta-ko-tourism-synth-cpt
rosetta-ko-tourism-synth-cpt
Korean tourism-domain data grounded in public tourism records — a cleaned CPT corpus plus source-grounded QA/preference/RLVR sets.
Synthetic data generated with the Qwen3.6-27B teacher model — part of the Rosetta-KO suite for the Rosetta Korean LLM (PoSTMEDIA).
Tourism Suite
Sibling datasets from the same pipeline (each a separate repo):
repo
format
rosetta-ko-tourism-synth-cpt (this repo)
continued-pretraining corpus (plain… See the full description on the dataset page: https://huggingface.co/datasets/PoSTMEDIA/rosetta-ko-tourism-synth-cpt.rosetta-ko-tourism-synth-rlvr
rosetta-ko-tourism-synth-rlvr
Korean tourism-domain data grounded in public tourism records — a cleaned CPT corpus plus source-grounded QA/preference/RLVR sets.
Synthetic data generated with the Qwen3.6-27B teacher model — part of the Rosetta-KO suite for the Rosetta Korean LLM (PoSTMEDIA).
Tourism Suite
Sibling datasets from the same pipeline (each a separate repo):
repo
format
rosetta-ko-tourism-synth-cpt
continued-pretraining corpus (plain text)… See the full description on the dataset page: https://huggingface.co/datasets/PoSTMEDIA/rosetta-ko-tourism-synth-rlvr.rose_code-Qwen3-1.7B-Pass8-Rollouts
rose_code rollouts — Qwen3-1.7B (pass@8)
Model rollouts on the rose_code test split, for the OPD coding pipeline.
Model / sampler: Qwen3-1.7B
Source prompts: CL-From-Nothing/rose_code test split — 408 competitive-programming questions (codeforces-style)
Sampling: 8 samples/question (pass@8) = 3264 records, temperature 0.7, max 16384 new tokens, max_model_len 32000
Rewards: DeepCoder code verifier (deepcoder_reward_fn.py) — 1.0 if the generated program passes all unit tests… See the full description on the dataset page: https://huggingface.co/datasets/CL-From-Nothing/rose_code-Qwen3-1.7B-Pass8-Rollouts.