datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
coding
Dataset Card for "livebench/coding"
LiveBench is a benchmark for LLMs designed with test set contamination and objective evaluation in mind. It has the following properties:
LiveBench is designed to limit potential contamination by releasing new questions monthly, as well as having questions based on recently-released datasets, arXiv papers, news articles, and IMDb movie synopses.
Each question has verifiable, objective ground-truth answers, allowing hard questions to be scored… See the full description on the dataset page: https://huggingface.co/datasets/livebench/coding.verifiable-coding-problems
SYNTHETIC-1
This is a subset of the task data used to construct SYNTHETIC-1. You can find the full collection here
Omni-Frontier-Distillation-SFT-Cyber-Coding-Med-dataset-collection
🧬 Omni-Frontier Collection
Cybersecurity · Coding · Math · Science · RSI Reasoning — one unified SFT package
A unified, deduplicated, fully-browsable distillation & SFT corpus — every row real, every row visible.
📖 Jump to
What's inside · 🔁 Aggregation audit · 🛡 Cybersecurity · 💻 Coding · 🏭 Distillation deep-dive · 🔁 RSI · 🧮 Math/Science/More · 🎓 Training guide · 🔎 Browsing · 🧹 Quality · 🗺 Roadmap · 📄 License… See the full description on the dataset page: https://huggingface.co/datasets/SHSLab/Omni-Frontier-Distillation-SFT-Cyber-Coding-Med-dataset-collection.verifiable-coding-problems-python
Dataset Card for Verifiable Coding Problems Python 10k
This dataset contains all Python problems from PrimeIntellect's verifiable-coding-problems dataset. We have formatted the verification_info and metadata columns to be proper dictionaries, but otherwise the data is the same. Please see their dataset for more details.
Omni-Frontier-Distillation-SFT-Cyber-Coding-Med-dataset-collection
🌌 Omni-Frontier Distillation SFT
The Definitive Evolution of Open-Source Distillation & Human-Crafted Expertise
Repository: Manusagents/Omni-Frontier-Distillation-SFT-Cyber-Coding-Med-dataset-collection
"The most comprehensive multi‑domain SFT corpus ever assembled — fusing 6.86 million cleaned distillation samples with 9.14 million human‑crafted expert examples across medical, cybersecurity, chemical, robotics, humanities, and more. 16 million… See the full description on the dataset page: https://huggingface.co/datasets/Manusagents/Omni-Frontier-Distillation-SFT-Cyber-Coding-Med-dataset-collection.Nemotron-RL-coding-competitive_coding
Dataset Description:
The Nemotron-RL-coding-competitive_coding dataset is a python-only, reasoning-based, synthetic dataset. It contains competitive coding style problems and their unit test cases. These questions and test cases are collected from CodeContests (deepmind/code_contests), and Open-R1 (open-r1/codeforces)
.
This dataset is released as part of NVIDIA NeMo Gym, a framework for building reinforcement learning environments to train large language models. NeMo Gym… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-RL-coding-competitive_coding.fable-5-coding-and-debugging-traces
Claude Fable 5 Agent Traces
2,380 TRAJECTORIES · 12,490 TRAINING ROWS · 14 MB PARQUET · 663 MB JSONL
Generated by moonshiner — an open harness for
distilling verified instruction-following, tool-use, and agentic coding traces.
Behavior-preserving instruction-following, tool-use, and agent trajectories
from Claude Fable 5 (anthropic/claude-fable-5). The category and row-share tables
below describe the actual mix seen during training rather than assuming a
particular task… See the full description on the dataset page: https://huggingface.co/datasets/DSFFGFG456/fable-5-coding-and-debugging-traces.kimi-k3-coding-and-debugging-traces
Kimi K3 Coding, Tool Use & Instruction Following Traces
582 TRAJECTORIES · 3,956 TRAINING ROWS · 3 MB PARQUET · 72 MB JSONL
Generated by moonshiner — an open harness for
distilling verified instruction-following, tool-use, and agentic coding traces.
Behavior-preserving instruction-following, tool-use, and agent trajectories
from Kimi K3 (moonshotai/kimi-k3). The category and row-share tables
below describe the actual mix seen during training rather than assuming a… See the full description on the dataset page: https://huggingface.co/datasets/greghavens/kimi-k3-coding-and-debugging-traces.web-research-coding-5m
Web Research + GitHub + Website Coding Dataset
Version: 1.1.0
Total examples: 5,000,000
Splits
train: 4,750,000
validation: 125,000
test: 125,000
Core capabilities
Web search
Web research
Evidence extraction
Fact verification
Multi-hop research
Multi-layer technical analysis
Architecture analysis
Root-cause analysis
Security analysis
Performance analysis
UX analysis
Design analysis
Refactoring
Code review
Debugging
Website coding
Design systems… See the full description on the dataset page: https://huggingface.co/datasets/Lelonthecodeur/web-research-coding-5m.Nemotron-RL-coding-quality-filtered
Nemotron coding — quality pool, revision 2
8,201 retained tasks from 16,083 upstream train rows (51.0%).
This is a static quality screen for Python standard-input/standard-output coding RL. It is not a reference-verified gold dataset.
All difficulty levels are eligible. There is no model-accuracy filter, rollout generation, rating cutoff, random subsampling, or 3,200-row cap.
Source: NVIDIA Nemotron-RL-coding-competitive_coding, revision 755d5910fc8646b385e3926eec08c152051cdc07.… See the full description on the dataset page: https://huggingface.co/datasets/hi-todayis-jh/Nemotron-RL-coding-quality-filtered.coding_agent_tracesWe release coding agent traces using Claude Code for
Opus
ISL, OSL, ISL_new counts
GPT-oss-120B
ISL, OSL, ISL_new counts and their raw texts
For Opus, only the locally saved files from the harness were used for analysis.
Coding agents take multiple turns to carry out a task from the input prompt. To analyze the token distribution, two models were selected: Anthropic's Opus and OpenAI's gpt-oss-120B. The input sequence length (ISL), output sequence length (OSL) and the uncached, new input… See the full description on the dataset page: https://huggingface.co/datasets/netpreme/coding_agent_traces.verifiable-coding-problems-python_decontaminated-testedverifiable-coding-problems-python_decontaminated-tested-shuffledglm-5.2-coding-and-debugging-traces
GLM 5.2 Agent Traces
207 TRAJECTORIES · 1,821 TRAINING ROWS · 1 MB PARQUET · 35 MB JSONL
Generated by moonshiner — an open harness for
distilling verified instruction-following, tool-use, and agentic coding traces.
Behavior-preserving instruction-following, tool-use, and agent trajectories
from GLM 5.2 (glm-5.2). The category and row-share tables
below describe the actual mix seen during training rather than assuming a
particular task domain.
This is an actively growing… See the full description on the dataset page: https://huggingface.co/datasets/greghavens/glm-5.2-coding-and-debugging-traces.verifiable-coding-problems-python_decontaminatedNemotron-LeetCode-coding-clean-3.2k
Nemotron + LeetCode Coding Clean 3.2k
3,200 distinct training problems, seed 42, intended for Python coding reinforcement learning. This is a training mix, not a held-out benchmark. It combines the pinned default train Parquet split of Nemotron-RL-coding-competitive_coding with the train JSONL of LeetCodeDataset.
Composition
Source
Questions
Selection
Nemotron / Codeforces
1,856
1000–1600, inclusive
Nemotron / AtCoder
366
300–2000 display difficulty… See the full description on the dataset page: https://huggingface.co/datasets/hi-todayis-jh/Nemotron-LeetCode-coding-clean-3.2k.variant_effect_coding
🧬 BioReasonIncentivizing Multimodal Biological Reasoning within a DNA-LLM Model
Variant Effect Coding Dataset
50,083 core variant entries from GPN-MSA study using ClinVar pathogenic variants and gnomAD benign variants (MAF>5%), split by chromosome (Chr 1-7,9-22,X,Y for train, Chr 8 for test) for pathogenic/benign classification.
Usage
from datasets import load_dataset
dataset = load_dataset("wanglab/variant_effect_coding")
example = dataset["train"][0]… See the full description on the dataset page: https://huggingface.co/datasets/wanglab/variant_effect_coding.Genomic_Benchmarks_demo_coding_vs_intergenomic_seqs
Dataset Card for "Genomic_Benchmarks_demo_coding_vs_intergenomic_seqs"
More Information needed
Omni-Frontier-Distillation-SFT-Cyber-security-Coding-dataset-collection-v2
🧬 Omni-Frontier Collection
Cybersecurity · Coding · Math · Science · RSI Reasoning — one unified SFT package
A unified, deduplicated, fully-browsable distillation & SFT corpus — every row real, every row visible.
📖 Jump to
What's inside · 🔁 Aggregation audit · 🛡 Cybersecurity · 💻 Coding · 🏭 Distillation deep-dive · 🔁 RSI · 🧮 Math/Science/More · 🎓 Training guide · 🔎 Browsing · 🧹 Quality · 🗺 Roadmap · 📄 License… See the full description on the dataset page: https://huggingface.co/datasets/Manusagents/Omni-Frontier-Distillation-SFT-Cyber-security-Coding-dataset-collection-v2.Organized_PreTrain_Coding_239kSI2CA-Training-TrajectoriesDataset Card for SI2CA-Training-Trajectories
[🌐 Website] •
[🤗 Dataset] •
[📜 Paper] •
[🐱 GitHub]
💡 Introduction
This dataset consists of 32,340 coding-agent trajectories generated by Qwen3.5-122B-A10B on the same 10,780 executable Python SWE tasks under the three trajectory-curation settings of Section 4.4 of the paper: standard sampling, full self-judgement, and an efficient discovered strategy found by the recursive self-improvement framework. Each task is… See the full description on the dataset page: https://huggingface.co/datasets/Self-Improving-Coding-Agents/SI2CA-Training-Trajectories.german-wikipedia-articleskimi-k3-coding-and-debugging-traces
Kimi K3 Coding, Tool Use & Instruction Following Traces
582 TRAJECTORIES · 3,956 TRAINING ROWS · 3 MB PARQUET · 72 MB JSONL
Generated by moonshiner — an open harness for
distilling verified instruction-following, tool-use, and agentic coding traces.
Behavior-preserving instruction-following, tool-use, and agent trajectories
from Kimi K3 (moonshotai/kimi-k3). The category and row-share tables
below describe the actual mix seen during training rather than assuming a… See the full description on the dataset page: https://huggingface.co/datasets/11-47/kimi-k3-coding-and-debugging-traces.guidellm-agentic-coding-trajectories
GuideLLM agentic coding trajectories
A sampled serving-load benchmark derived from Thoughtworks agentic-coding-trajectories, for GuideLLM and an OpenAI-compatible /v1/chat/completions endpoint. There are 630 rows representing 481 unique source sessions, across the same 8turn, 24turn, and 48turn configurations as the earlier version.
The configuration names now refer to original logical steps, not always HTTP request counts. Native tool steps expand into a tool-call request and a… See the full description on the dataset page: https://huggingface.co/datasets/zetomatoz/guidellm-agentic-coding-trajectories.agentic-coding-trajectories
agentic-coding-trajectories
A unified, tokenized corpus of 15,000 multi-turn agentic-coding sessions (618K turns, 41 turns/session avg) drawn from three publicly-released upstream datasets. Built for benchmarking LLM serving systems on realistic multi-turn coding-agent workloads.
Why this exists
Most LLM serving benchmarks use single-shot prompts. Real coding agents work in long multi-turn loops where each turn appends to a growing prompt. This corpus captures that shape… See the full description on the dataset page: https://huggingface.co/datasets/thoughtworks/agentic-coding-trajectories.agentic-coding-trajectories-grok46
Agentic Coding Trajectories (Grok 4.6)
Rights & intended use: public research corpus, not training data.
Hosted frontier-model outputs are research-only inputs under project policy
(synthetic-factory#161):
intended_use: research_only, project_training_policy: blocked. Not
training data for any model-weight update. Machine-readable record:
rights.json. License:
Synthetic Factory Research-Only License v1.0 (license: other, see LICENSE) (non-commercial).
Release status: the raw… See the full description on the dataset page: https://huggingface.co/datasets/rmems/agentic-coding-trajectories-grok46.coding_valagentic-coding-trajectories
Agentic Coding Trajectories
Rights & intended use: legacy public research corpus / portfolio
artifact. Hosted frontier-model outputs are research-only inputs under
project policy (synthetic-factory#161):
intended_use: research_only, project_training_policy: blocked. Not
training data for any model-weight update. Machine-readable record:
rights.json.
Release status: The raw, uncurated coding-episode payload is published under
data/raw/. It is available for inspection and… See the full description on the dataset page: https://huggingface.co/datasets/rmems/agentic-coding-trajectories.coding-agent-security-benchmark
Coding Agent Security Benchmark
A benchmark for evaluating whether an LLM can correctly identify security
violations in the behavior of an autonomous coding agent - spanning
dangerous shell commands, credential leakage, prompt injection, supply-chain
risk, privacy leaks, and more.
Each row is a single message sampled from a coding-agent session (a user
instruction, a tool call the agent issued, a tool's response, or the agent's
own output) paired with a ground-truth security… See the full description on the dataset page: https://huggingface.co/datasets/rogue-security/coding-agent-security-benchmark.glm-5.2-coding-and-debugging-traces
GLM 5.2 Agent Traces
207 TRAJECTORIES · 1,821 TRAINING ROWS · 1 MB PARQUET · 35 MB JSONL
Generated by moonshiner — an open harness for
distilling verified instruction-following, tool-use, and agentic coding traces.
Behavior-preserving instruction-following, tool-use, and agent trajectories
from GLM 5.2 (glm-5.2). The category and row-share tables
below describe the actual mix seen during training rather than assuming a
particular task domain.
This is an actively growing… See the full description on the dataset page: https://huggingface.co/datasets/11-47/glm-5.2-coding-and-debugging-traces.