datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
SWE-smith
SWE-smith Dataset
Code
•
Paper
•
Site
[12/14/2025] NOTE: We will no longer actively update this dataset.
While this dataset is still functional and usable, we recommend you use the `SWE-bench/SWE-smith-[lang]` datasets.
For better maintainability and ease-of-use, we are maintaining language-specific datasets in lieu of this mono-repo.
The SWE-smith Dataset is a training dataset of 50137 task instances from 128 GitHub repositories, collected using the SWE-smith toolkit.… See the full description on the dataset page: https://huggingface.co/datasets/SWE-bench/SWE-smith.SWE-smith-py
SWE-smith Dataset
Code
•
Paper
•
Site
As of 12/14/2025, SWE-smith: Python contains 50908 task instances from 131 GitHub repositories
The SWE-smith Dataset is the largest open source dataset for training software engineering agents.
All SWE-smith task instances come with an executable environment.
To learn more about how to use this dataset to train Language Models for Software Engineering, please refer to the documentation.
SWE-smith-trajectories
SWE-smith Trajectories
Code
•
Paper
•
Site
This dataset contains the 5017 trajectories we fine-tuned Qwen 2.5 Coder Instruct on, leading to
SWE-agent-LM-32B, a coding LM agent that
achieve 40.2% on SWE-bench Verified (no verifiers or multiple rollouts, just 1 attempt per instance).
Trajectories were generated by running SWE-agent + Claude 3.7 Sonnet on task instances from
the SWE-smith dataset.
minty-astro-ph
MINT-1T ArXiv Astro-ph
An astronomy-focused subset of mlfoundations/MINT-1T-ArXiv, filtered to include only papers from the astro-ph arXiv category (including cross-listed papers).
Overview
Papers
~845k
Total size
~804 GB
Format
WebDataset tar shards
Shards
287 (astro-ph-00000.tar to astro-ph-00286.tar)
Shard size
~3 GB each
Source
MINT-1T (Awadalla et al., 2024)
Data Format
Each tar shard contains paired files per paper:… See the full description on the dataset page: https://huggingface.co/datasets/Smith42/minty-astro-ph.SWE-smith-go
SWE-smith Dataset
Code
•
Paper
•
Site
As of 12/14/202, SWE-smith: Golang contains 8212 task instances from 87 GitHub repositories
The SWE-smith Dataset is the largest open source dataset for training software engineering agents.
All SWE-smith task instances come with an executable environment.
To learn more about how to use this dataset to train Language Models for Software Engineering, please refer to the documentation.
PubChem-124M-SMILES-SELFIES-InChI-IUPAC
PubChem-124M-Canonicalized-SELFIES-InChI-IUPAC
Dataset Summary
This dataset contains ~124 million chemical structures sourced from PubChem (as of Jan 2026), processed into a clean, machine-learning-ready Parquet format.
Unlike raw XML/JSON dumps or standard CSVs, this dataset provides a unified, tabular structure that joins multiple chemical identifiers and descriptors into a single sharded resource:
SMILES: Raw and RDKit-Canonicalized.
SELFIES: Pre-computed 100% robust… See the full description on the dataset page: https://huggingface.co/datasets/hheiden/PubChem-124M-SMILES-SELFIES-InChI-IUPAC.SWE-Smith
A extended version of the original SWE-smith-py dataset with more problem descriptions!
PubChem-124M-SMILES-SELFIES-InChI-IUPAC
PubChem-124M-Canonicalized-SELFIES-InChI-IUPAC
Dataset Summary
This dataset contains ~124 million chemical structures sourced from PubChem (as of Jan 2026), processed into a clean, machine-learning-ready Parquet format.
Unlike raw XML/JSON dumps or standard CSVs, this dataset provides a unified, tabular structure that joins multiple chemical identifiers and descriptors into a single sharded resource:
SMILES: Raw and RDKit-Canonicalized.
SELFIES: Pre-computed… See the full description on the dataset page: https://huggingface.co/datasets/Bilsteen/PubChem-124M-SMILES-SELFIES-InChI-IUPAC.swe_smith_rebenchv2_5136
SWE-smith + SWE-rebench V2 5136 Mix
This dataset is the swe_smith_rebenchv2_5136 training mix used by the rLLM SWE training scripts. It combines filtered SWE-smith trajectory tasks with sampled SWE-rebench V2 tasks so future training jobs can pull the prepared parquet directly instead of regenerating it with the long preparation script.
Contents
data/train.parquet: the canonical rLLM task rows, 5,136 examples.
rllm_verl/train.parquet: the rLLM DatasetRegistry… See the full description on the dataset page: https://huggingface.co/datasets/JWei05/swe_smith_rebenchv2_5136.SWE-smith-trajectories
SWE-smith Trajectories
Code
•
Paper
•
Site
This dataset contains the 5017 trajectories we fine-tuned Qwen 2.5 Coder Instruct on, leading to
SWE-agent-LM-32B, a coding LM agent that
achieve 40.2% on SWE-bench Verified (no verifiers or multiple rollouts, just 1 attempt per instance).
Trajectories were generated by running SWE-agent + Claude 3.7 Sonnet on task instances from
the SWE-smith dataset.
smithsonian-data
Smithsonian Open Access Data
Pre-processed data dumps from the Smithsonian Open Access initiative, covering millions of objects across Smithsonian Institution museums and archives.
What is this?
The Smithsonian publishes their Open Access metadata on S3, but the raw data is split across 255 individual .txt files per unit. This dataset consolidates each unit's data into a single .jsonl.gz file for easier downloading and processing.
Files
Each file corresponds to… See the full description on the dataset page: https://huggingface.co/datasets/museado/smithsonian-data.ascl-code
ASCL Astronomy Source Code
The Astrophysics Source Code Library (ASCL) is a curated registry of
source code used in astronomy and astrophysics research. This dataset contains source files
extracted from ASCL-listed repositories, paired with catalog metadata.
Dataset Structure
Manifest (manifest.parquet)
One row per ASCL catalog entry with the following fields:
Field
Description
ascl_id
ASCL identifier (e.g., [ascl:2306.019])
title
Software title… See the full description on the dataset page: https://huggingface.co/datasets/Smith42/ascl-code.SWE-smith-code-search
SWE-smith Code Search
SWE-smith Code Search is a code-localization training dataset derived from
SWE-bench/SWE-smith-py.
It converts each eligible SWE-smith mutation patch into structured file,
class/module, and function/method targets. The release contains 39,291 examples
from 131 Python repositories and is intended for training and evaluating coding
agents that must locate relevant code before editing it.
中文简介：本数据集从 SWE-smith Python 原始任务中提取文件、类/模块、函数/方法三级
代码定位标签，可直接用于 Code… See the full description on the dataset page: https://huggingface.co/datasets/LeeXugar/SWE-smith-code-search.Tool-Calling-Dataset-UIGEN-X
Tool Calling Dataset for UIGEN-X
This is a cleaned version of the interstellarninja/hermes_reasoning_tool_use dataset, specifically prepared for training with Axolotl.
Usage with Axolotl
datasets:
- path: smirki/Tool-Calling-Dataset-UIGEN-X
type: chat_template
field_messages: conversations
message_property_mappings:
role: from
content: value
split: train
Example Structure
{
"conversations": [
{
"from": "system"… See the full description on the dataset page: https://huggingface.co/datasets/smirki/Tool-Calling-Dataset-UIGEN-X.PubChem-124M-SMILES-SELFIES-InChI-IUPAC
PubChem-124M-Canonicalized-SELFIES-InChI-IUPAC
Dataset Summary
This dataset contains ~124 million chemical structures sourced from PubChem (as of Jan 2026), processed into a clean, machine-learning-ready Parquet format.
Unlike raw XML/JSON dumps or standard CSVs, this dataset provides a unified, tabular structure that joins multiple chemical identifiers and descriptors into a single sharded resource:
SMILES: Raw and RDKit-Canonicalized.
SELFIES: Pre-computed 100% robust… See the full description on the dataset page: https://huggingface.co/datasets/th-laurel/PubChem-124M-SMILES-SELFIES-InChI-IUPAC.swe-smith-frozen-trajectories-openai
SWE-Smith Frozen Trajectories — OpenAI Wire Format
This dataset is the OpenAI chat-completions wire-format release of
reflectio/swe-smith-frozen-trajectories,
derived from the tool split of
SWE-bench/SWE-smith-trajectories.
It is a serving-performance workload for realistic multi-turn coding-agent
histories. It can be used to measure request throughput, input/output token
throughput, TTFT, TPOT, streaming behavior, and prefix-cache reuse. It is not
a coding-correctness… See the full description on the dataset page: https://huggingface.co/datasets/reflectio/swe-smith-frozen-trajectories-openai.swe-smith-frozen-trajectories
SWE-Smith Frozen Trajectories
This dataset is a serving-performance workload derived from the tool split of
SWE-bench/SWE-smith-trajectories.
It is designed for measuring throughput, request rate, time to first token,
inter-token latency, and prefix-cache behavior with realistic multi-turn coding
agent histories.
It is not a coding-correctness benchmark. The tested model's responses are
not executed or scored.
Processing
Keep trajectories generated by… See the full description on the dataset page: https://huggingface.co/datasets/reflectio/swe-smith-frozen-trajectories.chemq3-molsim-sft-smiles
ECFP4 Molecular Pairs Dataset
A dataset of molecular pairs with ECFP4 Dice similarity scores uniformly sampled across a target range, using FAISS for efficient similarity search.
This pipeline intended to generate a high-quality dataset of molecular pairs for similarity-based learning, balancing chemical diversity, computational efficiency, and target similarity distribution.
Specially designed to retain only pairs with 0.5 ≤ Dice(MACCS) ≤ 0.95—a targeted range for supervised… See the full description on the dataset page: https://huggingface.co/datasets/gbyuvd/chemq3-molsim-sft-smiles.Agentic-Coding-Tessa
Agentic Coding Dataset for Tessa
A comprehensive dataset for training coding agents with tool-use, reasoning, and software engineering capabilities.
Dataset Composition
This dataset combines multiple high-quality sources:
hermes_reasoning (20.0%): Tool-use and reasoning dataset - interstellarninja/hermes_reasoning_tool_use
search_arena (15.0%): Search and retrieval tasks - lmarena-ai/search-arena-24k
arena_human_pref (15.0%): Human preference data for alignment -… See the full description on the dataset page: https://huggingface.co/datasets/smirki/Agentic-Coding-Tessa.SWE-smith-trajectories
SWE-smith Trajectories
Code
•
Paper
•
Site
This dataset contains the 5017 trajectories we fine-tuned Qwen 2.5 Coder Instruct on, leading to
SWE-agent-LM-32B, a coding LM agent that
achieve 40.2% on SWE-bench Verified (no verifiers or multiple rollouts, just 1 attempt per instance).
Trajectories were generated by running SWE-agent + Claude 3.7 Sonnet on task instances from
the SWE-smith dataset.
swe-smith-py-glm5.2-tiny
swe-smith-py-glm5.2-tiny
SWE-smith Python trajectories collected with GLM-5.2, filtered to evaluation-resolved instances, and exported as TRL-style chat messages for supervised fine-tuning (tool calling + <think> reasoning).
Intended publish path: pyromind/swe-smith-py-glm5.2-tiny.
Summary
Item
Value
Model
openai/glm-5.2-fp8
Agent
mini-swe-agent (2.4.5)
Environment
Docker (/testbed)
Repo / task family
oauthlib SWE-smith mutations
Eval run… See the full description on the dataset page: https://huggingface.co/datasets/pyromind/swe-smith-py-glm5.2-tiny.counseling-smile
SMILE — Counseling Dialogues (processed)
大规模中文通用心理支持对话(SMILE, 5.5 万段),适合通用 SFT。
本仓库是 counselor_agent 项目中,经统一预处理器落地到 dataset/processed/ 的
SMILE 数据集。所有记录采用统一 schema(case_id / source / lang /
messages[] + 各数据集特有的可选标注 / profile)。
规模
smile: 55,124 dialogues, 627,674 turns (avg 11.39)
文件
文件
类型
条数
大小
all.jsonl
dialogue
55,124
174.0 MB
all.turns.jsonl
turns
627,674
262.8 MB
dialogue:对话级,一行 = 一段完整对话(messages 为完整对话轮列表)。
turns:轮次级,一行 = 当前轮… See the full description on the dataset page: https://huggingface.co/datasets/XuShihao6715/counseling-smile.taboo-smile
taboo-smile
This dataset contains conversational data in JSONL format, suitable for Supervised Fine-Tuning (SFT).
Usage
from datasets import load_dataset
# Load the dataset
dataset = load_dataset("bcywinski/taboo-smile")
Format
The dataset is in JSONL format where each line contains a conversation record suitable for training chat models.
Sam-1-large-identity-and-safetya dataset teaching the newest sam 1 large LLM its identity
SMILEContributors: Baisakhi Sarkar, Chakita Muttaraju, Xinyi (Cindy) Lyu
Introduction
SMILE (Synthetic Multi-turn Interactions for Learning Ethics) is a synthetic dataset consisting of multi-turn, text + image conversations between a human
and an AI agent focusing on improving multimodal model performance on the 3Hs (Helpful, Honest, Harmless) as well as for implementing necessary safety
and privacy restrictions such as not identifying persons from a given image.
Dataset… See the full description on the dataset page: https://huggingface.co/datasets/Chakita/SMILE.bioactives-naturals-smiles-molgen
Valid Bioactives and Natural Product SMILES
~2.7M valid SMILES built and curated from ChemBL34 (Zdrazil et al. 2023), COCONUTDB (Sorokina et al. 2021), and Supernatural3 (Gallo et al. 2023) dataset.
Curated by: gbyuvd
References
BibTeX
COCONUTDB
@article{sorokina2021coconut,
title={COCONUT online: Collection of Open Natural Products database},
author={Sorokina, Maria and Merseburger, Peter and Rajan, Kohulan and Yirik, Mehmet Aziz and Steinbeck… See the full description on the dataset page: https://huggingface.co/datasets/gbyuvd/bioactives-naturals-smiles-molgen.paper-tech-data-text
Dataset: paper-tech-data-text
Dataset hasil olahan untuk proyek buku LLM.
combined-sft-dataset
Combined SFT Dataset
Unified dataset combining multiple sources for LLaDA2 SFT training.
Format
JSONL with messages (array of {role, content} objects) and source (string) per row.
The last message in every row has role: "assistant".
Sources
Source
Description
opus-4.6-reasoning-3000x
Opus 4.6 reasoning (filtered)
claude-4.5-opus-reasoning-250x
Claude 4.5 Opus high reasoning
openresearcher
OpenResearcher research QA
toolmind-web-qa
ToolMind Web… See the full description on the dataset page: https://huggingface.co/datasets/smirki/combined-sft-dataset.BiasDPO_ID
Dataset: BiasDPO_ID
Dataset hasil olahan untuk proyek buku LLM.