datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
codeparrot-clean
CodeParrot 🦜 Dataset Cleaned
What is it?
A dataset of Python files from Github. This is the deduplicated version of the codeparrot.
Processing
The original dataset contains a lot of duplicated and noisy data. Therefore, the dataset was cleaned with the following steps:
Deduplication
Remove exact matches
Filtering
Average line length < 100
Maximum line length < 1000
Alpha numeric characters fraction > 0.25
Remove auto-generated files (keyword search)
For… See the full description on the dataset page: https://huggingface.co/datasets/codeparrot/codeparrot-clean.CodeFeedback-Filtered-Instruction OpenCodeInterpreter: Integrating Code Generation with Execution and Refinement
[🏠Homepage]
|
[🛠️Code]
OpenCodeInterpreter
OpenCodeInterpreter is a family of open-source code generation systems designed to bridge the gap between large language models and advanced proprietary systems like the GPT-4 Code Interpreter. It significantly advances code generation capabilities by integrating execution and iterative refinement functionalities.
For further information and… See the full description on the dataset page: https://huggingface.co/datasets/m-a-p/CodeFeedback-Filtered-Instruction.CodeAlpaca-20kEvol-Instruct-Code-80k-v1Open Source Implementation of Evol-Instruct-Code as described in the WizardCoder Paper.
Code for the intruction generation can be found on Github as Evol-Teacher.
python-codes-25k
License
MIT
This is a Cleaned Python Dataset Covering 25,000 Instructional Tasks
Overview
The dataset has 4 key features (fields): instruction, input, output, and text.It's a rich source for Python codes, tasks, and extends into behavioral aspects.
Dataset Statistics
Total Entries: 24,813
Unique Instructions: 24,580
Unique Inputs: 3,666
Unique Outputs: 24,581
Unique Texts: 24,813
Average Tokens per example: 508
Features… See the full description on the dataset page: https://huggingface.co/datasets/flytech/python-codes-25k.evol-codealpaca-v1
Evolved codealpaca
Updates:
2023/08/26 - Filtered results now only contain pure english instruction and removed any mentioned of trained by OAI response
Median sequence length : 471
We employed a methodology similar to that of WizardCoder, with the exception that ours is open-source. We used the gpt-4-0314 and gpt-4-0613 models to augment and answer each response, with the bulk of generation handled by gpt-4-0314.
The aim of this dataset is twofold: firstly, to facilitate the… See the full description on the dataset page: https://huggingface.co/datasets/theblackcat102/evol-codealpaca-v1.Java-Code-Large-text-onlyJava-Code-Large
Java-Code-Large is a large-scale corpus of publicly available Java source code comprising more than 15 million java codes. The dataset is designed to support research in large language model (LLM) pretraining, code intelligence, software engineering automation, and program analysis.
By providing a high-volume, language-specific corpus, Java-Code-Large enables systematic experimentation in Java-focused model training, domain adaptation, and downstream code understanding tasks.… See the full description on the dataset page: https://huggingface.co/datasets/XEUIPR/Java-Code-Large-text-only.code_generation
LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code
🏠 Home Page •
💻 GitHub Repository •
🏆 Leaderboard •
LiveCodeBench is a "live" updating benchmark for holistically evaluating code related capabilities of LLMs.
Particularly, it evaluates LLMs across a range of capabilties including code generation, self-repair, test output prediction, and code execution.
This is the code generation scenario of LiveCodeBench. It is also… See the full description on the dataset page: https://huggingface.co/datasets/livecodebench/code_generation.swallow-code-v2
SwallowCode-v2
Resources
📑 arXiv: Read our paper for detailed methodology and results at arXiv:2505.02881.
🤗 Sister Dataset: Discover SwallowMath-v2, our companion dataset for mathematical reasoning.
💻 What is it?
SwallowCode-v1 was a high-quality Python code dataset generated through an LLM-based rewriting pipeline.
However, it had two significant limitations:
(1) it was distributed under the Llama 3.3 Community License, and
(2) its size was limited to… See the full description on the dataset page: https://huggingface.co/datasets/tokyotech-llm/swallow-code-v2.claude-fable-5-claude-code
claude-fable-5 Agent Traces
It's worth noting that our team was working with Glint-Research to collect as much fable data as possible.
These are just the anonymized raw traces of both of our teams combined. This means that Glint-Research/Fable-5-traces was created from formatting and splitting up this same dataset. If you use one for your tune, don't use the other (it's the same exact data).
For training on this dataset I recommend using the teich package to convert to openai… See the full description on the dataset page: https://huggingface.co/datasets/armand0e/claude-fable-5-claude-code.sounio-code-examples
Sounio Curated Code Examples
Curated compile-clean .sio examples for training and evaluating code models on
Sounio, a self-hosted systems and scientific programming language for epistemic
computing, uncertainty propagation, and algebraic effects.
This directory is the Cx-1 expansion lane for
chiuratto-AIgourakis/sounio-code-examples.
Current batch
Examples: 5,000
Metadata files: 5,000
Compiler gate: bin/souc check pass rate 5,000/5,000
Utility layer: 5,000… See the full description on the dataset page: https://huggingface.co/datasets/chiuratto-AIgourakis/sounio-code-examples.style-dpo
gijl style dataset (multi-type)
Generated by meta-models/Muse-Glimmer-30B through a tool-using scouting loop over real sources (Stack Exchange, GitHub, OSV, Hacker News, arXiv, Wikipedia, web). Splits are a deterministic hash of the record id (90/5/5); derived records inherit their parent's split. Synthetic, model-written, not human-verified. Every rejected response is intentionally poor and must never be used as an example of good behavior.
config
folder
train
validation… See the full description on the dataset page: https://huggingface.co/datasets/codex-automatus/style-dpo.code_instructions_122k_alpaca_stylecodeparrot-clean-train
CodeParrot 🦜 Dataset Cleaned (train)
Train split of CodeParrot 🦜 Dataset Cleaned.
Dataset structure
DatasetDict({
train: Dataset({
features: ['repo_name', 'path', 'copies', 'size', 'content', 'license', 'hash', 'line_mean', 'line_max', 'alpha_frac', 'autogenerated'],
num_rows: 5300000
})
})
JavaScript-Code-LargeJavaScript-Code-Large
JavaScript-Code-Large is a large-scale corpus of JavaScript source code comprising around 5 million JavaScript files. The dataset is designed to support research in large language model (LLM) pretraining, code intelligence, software engineering automation, and program analysis for the JavaScript ecosystem.
By providing a high-volume, language-specific corpus, JavaScript-Code-Large enables systematic experimentation in JavaScript-focused model training, domain adaptation… See the full description on the dataset page: https://huggingface.co/datasets/ajibawa-2023/JavaScript-Code-Large.mimo-claude-code-traces-1k
MIMO Claude Code Traces
MIMO Claude Code Traces is a collection of coding-agent trajectories in a Claude Code-style environment. Each record contains a user coding task, the full multi-turn message trace, available tool schemas, assistant reasoning fields, tool calls, tool outputs, and metadata such as model name, category, duration, cost, token usage, and whether the trace used tools.
The traces were generated with mimo-v2.5-pro, MiMo's most capable model at the time of… See the full description on the dataset page: https://huggingface.co/datasets/choucsan/mimo-claude-code-traces-1k.Code-Feedback OpenCodeInterpreter: Integrating Code Generation with Execution and Refinement
[🏠Homepage]
|
[🛠️Code]
Introduction
OpenCodeInterpreter is a family of open-source code generation systems designed to bridge the gap between large language models and advanced proprietary systems like the GPT-4 Code Interpreter. It significantly advances code generation capabilities by integrating execution and iterative refinement functionalities.
For further information and related… See the full description on the dataset page: https://huggingface.co/datasets/m-a-p/Code-Feedback.R1-Code-Interpreter-Data
R1-Code-Interpreter: Training LLMs to Reason with Code via Supervised and Reinforcement Learning
Our code is based on Llama-factory/VeRL/Search-R1 for the SFT and RL training and SymBench/BIG-Bench-Hard/reasoning-gym for datasets/benchmarks of reasoning/planning tasks.
📝 Introduction
R1-Code-Interpreter is the first framework to train LLMs for step-by-step code reasoning using multi-turn supervised fine-tuning and reinforcement learning. By curating 144 diverse… See the full description on the dataset page: https://huggingface.co/datasets/yongchao98/R1-Code-Interpreter-Data.voice-code-bench
VoiceCodeBench
VoiceCodeBench is a test-only benchmark for evaluating whether automatic
speech recognition (ASR) systems preserve exact structured values in English
workplace speech.
Paper: VoiceCodeBench: Evaluating Exact Structured-Token Recovery in Automatic Speech Recognition
The benchmark targets cases where a transcript is software input: callback
numbers, email addresses, command-line flags, file paths, URLs, account
identifiers, dates, measurements, and similar values… See the full description on the dataset page: https://huggingface.co/datasets/besimple-ai/voice-code-bench.glaive-code-assistant-v3
Glaive-code-assistant-v3
Glaive-code-assistant-v3 is a dataset of ~1M code problems and solutions generated using Glaive’s synthetic data generation platform.
This is built on top of the previous version of the dataset that can be found here. This already includes v1 and v2 of the dataset.
To report any problems or suggestions in the data, join the Glaive discord
GPT-5.5-CodexThis dataset was generated using teich by TeichAI
GPT-5.5 Agent traces
This directory contains raw agent trace files generated by teich.
JSONL files: 317
Model metadata: gpt-5.5
Training-ready tools
Generated agent traces carry configured or recovered tool schemas so tools remain available for training even when a session did not call them.
Native Claude Code imports recover schemas for Claude Code and Claude Desktop built-ins, plus conservative name-derived MCP… See the full description on the dataset page: https://huggingface.co/datasets/AletheiaResearch/GPT-5.5-Codex.Code-eXtraClean Low Comment Rich Code this is High Quality code with Less Comments , More logic code that is dedented a good amount, to make faster coding agents that code without taking your whole hard-drive!Also HuggingFace doesn't want to show this but it is actualy ~2.9B tokens because we condensed it into the best for AI training decommentatioon and dedent.Example Code snippets:
os = require('os')
fs = require('fs')
program = require('commander')
worker = require('../lib/worker')
program… See the full description on the dataset page: https://huggingface.co/datasets/ZyperAI/Code-eXtra.Klear-CodeTestcodeparrot-ds-trainCode_Vulnerability_Security_DPO
Secure-code dataset: correction and audit notice (2026-10-06)
The chosen and rejected fields are original model-generated preference labels. They are not verified secure/insecure classifications. Do not use these labels as security ground truth or use the dataset as a validated secure-code training or evaluation set without your own contextual validation.
The historical description below overstates security, optimization, realism and validation. Those claims are superseded by… See the full description on the dataset page: https://huggingface.co/datasets/CyberNative/Code_Vulnerability_Security_DPO.glaive-code-assistant
Glaive-code-assistant
Glaive-code-assistant is a dataset of ~140k code problems and solutions generated using Glaive’s synthetic data generation platform.
The data is intended to be used to make models act as code assistants, and so the data is structured in a QA format where the questions are worded similar to how real users will ask code related questions.
The data has ~60% python samples.
To report any problems or suggestions in the data, join the Glaive discord
github-jupyter
GitHub Jupyter Dataset
Dataset Description
The dataset was extracted from Jupyter Notebooks on BigQuery.
Licenses
Each example has the license of its associated repository. There are in total 15 licenses:
[
'mit',
'apache-2.0',
'gpl-3.0',
'gpl-2.0',
'bsd-3-clause',
'agpl-3.0',
'lgpl-3.0',
'lgpl-2.1',
'bsd-2-clause',
'cc0-1.0',
'epl-1.0',
'mpl-2.0',
'unlicense',
'isc',
'artistic-2.0'
]
math-code-science-deepseek-r1-en
R1 Dataset Collection
Aggregated high-quality English prompts and model-generated responses from DeepSeek R1 and DeepSeek R1-0528.
Dataset Summary
The R1 Dataset Collection combines multiple public DeepSeek-generated instruction-response corpora into a single, cleaned, English-only JSONL file. Each example consists of a <|user|> prompt and a <|assistant|> response in one "text" field. This release includes:
~21,000 examples from the DeepSeek-R1-0528 Distilled Custom… See the full description on the dataset page: https://huggingface.co/datasets/Hugodonotexit/math-code-science-deepseek-r1-en.Cpp-Code-LargeCpp-Code-Large
Cpp-Code-Large is a large-scale corpus of C++ source code comprising more than 5 million lines of C++ code. The dataset is designed to support research in large language model (LLM) pretraining, code intelligence, software engineering automation, and static program analysis for the C++ ecosystem.
By providing a high-volume, language-specific corpus, Cpp-Code-Large enables systematic experimentation in C++-focused model training, domain adaptation, and downstream code… See the full description on the dataset page: https://huggingface.co/datasets/ajibawa-2023/Cpp-Code-Large.glaive-code-assistant-v2
Glaive-code-assistant-v2
Glaive-code-assistant-v2 is a dataset of ~215k code problems and solutions generated using Glaive’s synthetic data generation platform.
This is built on top of the previous version of the dataset that can be found here
To report any problems or suggestions in the data, join the Glaive discord