datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
Nemotron-Terminal-Corpus
Terminal-Corpus: Large-Scale SFT Dataset for Terminal Agents
Terminal-Corpus is a large-scale Supervised Fine-Tuning (SFT) dataset designed to scale the terminal interaction capabilities of Large Language Models (LLMs). Developed by NVIDIA, this dataset was built using the Terminal-Task-Gen pipeline, which combines dataset adaptation with synthetic task generation across diverse domains.
🚀 Key Results & Performance
The high-quality trajectories in Terminal-Corpus enable… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-Terminal-Corpus.Nemotron-Terminal-Synthetic-Tasks
Terminal-Corpus: Task Structure Specification
This repository contains the skill-based synthetic tasks within the Terminal-Corpus. These tasks are designed to evaluate and train autonomous agents in realistic Linux terminal environments.
🏗️ Task Anatomy
Each task is contained within a dedicated directory and follows a strict four-component architecture:
1. Instruction (instruction.md)
Purpose: Provides the natural language description of the objective.… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-Terminal-Synthetic-Tasks.Nemotron-AIQ-Agentic-Safety-Dataset-1.0
Nemotron-AIQ Agentic Safety Dataset
Dataset Summary
Nemotron-AIQ-Agentic-Safety-Dataset is a comprehensive dataset that captures a broad range of novel safety and security contextual risks that can emerge within agentic systems. It highlights the robustness of NVIDIA's open model, llama-3.3-nemotron-super-49b-v1, when deployed as a research assistant inside AIQ, demonstrating its ability to handle a diverse spectrum of agentic safety and security challenges. The dataset… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-AIQ-Agentic-Safety-Dataset-1.0.Nemotron-SpecializedDomains-Finance-v1
Dataset Description
Nemotron-SpecializedDomains-Finance is a large-scale synthetic financial question-answering dataset designed to improve LLM performance on specialized financial reasoning and document comprehension tasks. The dataset comprises 326K+ high-quality Q&A pairs generated from SEC filings of S&P 500 companies spanning 2019-2024.
This dataset is ready for commercial use.
Overview
The dataset leverages template-based Synthetic Data Generation (SDG) to… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-SpecializedDomains-Finance-v1.Nemotron-CrossThink
Nemotron-CrossThink: Scaling Self-Learning beyond Math Reasoning
Author: Syeda Nahida Akter, Shrimai Prabhumoye, Matvei Novikov, Seungju Han, Ying Lin, Evelina Bakhturina, Eric Nyberg, Yejin Choi,
Mostofa Patwary, Mohammad Shoeybi, Bryan Catanzaro
[Paper][Blog]
Dataset Description
Nemotron-CrossThink is a multi-domain reinforcement learning (RL) dataset designed to improve general-purpose
and mathematical reasoning in large language models (LLMs).
The dataset… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-CrossThink.Nemotron-RL-ARC-AGI-v1
Dataset Description:
Nemotron-RL-ARC-AGI-v1 is a reinforcement-learning (RL) gym environment dataset of single-step ARC-AGI puzzle prompts intended for RL post-training of large language models. Each row corresponds to one ARC puzzle (a set of (input grid, output grid) demonstration pairs plus a single test input grid) rendered as a text prompt; reward is binary (1.0 / 0.0) determined by exact-match comparison against the ground-truth output grid. No LLM judge is used, no… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-RL-ARC-AGI-v1.Nemotron-RL-litmus-bench-v0.1
Dataset Description:
Litmus-Bench v0.1 is an open dataset for training and evaluating chemical reasoning in language models. It includes 5,232 training questions and 482 test questions, each in short-answer format and was created from the ChEMBL dataset with RDKit descriptors requiring short answers. The dataset is for RL training.
This dataset is released as part of NVIDIA NeMo-Gym, an open-source library within the NVIDIA NeMo framework, designed for large-scale, verifiable… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-RL-litmus-bench-v0.1.Medical-Reasoning-SFT-Nemotron-Nano-30B
Medical-Reasoning-SFT-Nemotron-Nano-30B
A large-scale medical reasoning dataset generated using nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16, containing over 444,000 samples with detailed chain-of-thought reasoning for medical and healthcare questions.
Dataset Overview
Metric
Value
Model
nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16
Total Samples
444,544
Samples with Reasoning
444,544 (100%)
Estimated Tokens
~1.01 Billion
Content Tokens
~808 Million… See the full description on the dataset page: https://huggingface.co/datasets/OpenMed/Medical-Reasoning-SFT-Nemotron-Nano-30B.Nemotron-Math-v2
Nemotron-Math-v2 (Converted)
This is a conversion of nvidia/Nemotron-Math-v2 to Parquet format, preserving the original split structure.
Processed with high-speed parallel streaming conversion.
muse12-nemo-agentic
Muse Spark 1.2 High-Reasoning NeMo Agentic Dataset
A reproducible, verified 24,000-row synthetic agentic dataset generated with Meta Muse Spark 1.2, NeMo Gym, and deterministic task-family verifiers.
The project is a quality-focused successor to r0b0tlab/deepseek-v4-pro-0813-agentic. It keeps rollout prompts separate from reference trajectories and offline-training views, records usage and provenance, and does not publish private chain-of-thought.
[!IMPORTANT]
Status:… See the full description on the dataset page: https://huggingface.co/datasets/r0b0tlab/muse12-nemo-agentic.Nemotron-AIQ-Agentic-Safety-Dataset-1.0
Nemotron-AIQ Agentic Safety Dataset
Dataset Summary
Nemotron-AIQ-Agentic-Safety-Dataset is a comprehensive dataset that captures a broad range of novel safety and security contextual risks that can emerge within agentic systems. It highlights the robustness of NVIDIA's open model, llama-3.3-nemotron-super-49b-v1, when deployed as a research assistant inside AIQ, demonstrating its ability to handle a diverse spectrum of agentic safety and security challenges. The dataset… See the full description on the dataset page: https://huggingface.co/datasets/yuqing1207/Nemotron-AIQ-Agentic-Safety-Dataset-1.0.nemotron-terminal-security
nemotron-terminal-security
Per-source partition of nvidia/Nemotron-Terminal-Corpus,
filtered to source == "security". The difficulty column preserves the original
easy / medium / mixed split (na for the dataset_adapters/* files, which
did not carry a difficulty label).
Partitioning scheme:
adapters_{code,math,swe} — rows from dataset_adapters/{code,math,swe}.parquet
{skill} (e.g. debugging, security, …) — rows from
synthetic_tasks/skill_based/{easy,medium… See the full description on the dataset page: https://huggingface.co/datasets/laion/nemotron-terminal-security.Nemotron-RL-QA-Abstention-v1Nemotron-RL-QA-Abstention-v1
License: cc-by-4.0
Language: en
Task Categories: reinforcement-learning, question-answering, text-generation
Tags: abstention, question-answering, hotpotqa, software-engineering, health, law, rl, rlvr
Configs: default train split at data/train.jsonl
Domain: multi-domain question answering, abstention
Modality: text
Capability Breakdown: Abstention-aware factoid question answering [100%]
Source: Hybrid: Automated, Manually Collected, Synthetic
Size Bin: <10K… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-RL-QA-Abstention-v1.nemotron-terminal-debugging
nemotron-terminal-debugging
Per-source partition of nvidia/Nemotron-Terminal-Corpus,
filtered to source == "debugging". The difficulty column preserves the original
easy / medium / mixed split (na for the dataset_adapters/* files, which
did not carry a difficulty label).
Partitioning scheme:
adapters_{code,math,swe} — rows from dataset_adapters/{code,math,swe}.parquet
{skill} (e.g. debugging, security, …) — rows from
synthetic_tasks/skill_based/{easy,medium… See the full description on the dataset page: https://huggingface.co/datasets/laion/nemotron-terminal-debugging.nemotron_terminal_filtered
Nemotron Terminal Filtered
An uncertainty-curated subset of NVIDIA's Nemotron-Terminal-Corpus (dataset_adapters split), selected for high-formation density for post-training NVIDIA-Nemotron-3-Super-120B-A12B-BF16.
Motivation
The full dataset_adapters split contains ~226k terminal execution trajectories. To curate a compact, high-value subset for post-training we score each sample by how hard the model finds it, using entropy as a proxy for uncertainty. The… See the full description on the dataset page: https://huggingface.co/datasets/locailabs/nemotron_terminal_filtered.nemotron-terminal-file_operations
nemotron-terminal-file_operations
Per-source partition of nvidia/Nemotron-Terminal-Corpus,
filtered to source == "file_operations". The difficulty column preserves the original
easy / medium / mixed split (na for the dataset_adapters/* files, which
did not carry a difficulty label).
Partitioning scheme:
adapters_{code,math,swe} — rows from dataset_adapters/{code,math,swe}.parquet
{skill} (e.g. debugging, security, …) — rows from
synthetic_tasks/skill_based/{easy,medium… See the full description on the dataset page: https://huggingface.co/datasets/laion/nemotron-terminal-file_operations.Nemotron-Terminal-Corpus
Terminal-Corpus: Large-Scale SFT Dataset for Terminal Agents
Terminal-Corpus is a large-scale Supervised Fine-Tuning (SFT) dataset designed to scale the terminal interaction capabilities of Large Language Models (LLMs). Developed by NVIDIA, this dataset was built using the Terminal-Task-Gen pipeline, which combines dataset adaptation with synthetic task generation across diverse domains.
🚀 Key Results & Performance
The high-quality trajectories in Terminal-Corpus enable… See the full description on the dataset page: https://huggingface.co/datasets/txchmechanicus/Nemotron-Terminal-Corpus.nemotron-terminal-data_querying
nemotron-terminal-data_querying
Per-source partition of nvidia/Nemotron-Terminal-Corpus,
filtered to source == "data_querying". The difficulty column preserves the original
easy / medium / mixed split (na for the dataset_adapters/* files, which
did not carry a difficulty label).
Partitioning scheme:
adapters_{code,math,swe} — rows from dataset_adapters/{code,math,swe}.parquet
{skill} (e.g. debugging, security, …) — rows from
synthetic_tasks/skill_based/{easy,medium… See the full description on the dataset page: https://huggingface.co/datasets/laion/nemotron-terminal-data_querying.prosat-nemotron-alice-in-wonderland-traces
The Adimension's Pro-SAT | Programmatic Solver-Augmented Traces for Alice in Wonderland Puzzles
Alice meets Nemotron to reason efficiently in Wonderland!
A fully programmatic (zero LLM calls) system that generates Chain-of-Thought reasoning traces for the Alice in Wonderland puzzles from the NVIDIA Nemotron Model Reasoning Challenge.
Overview
The prosat_traces_generator.py script reverse-engineers deterministic transformation rules from puzzle… See the full description on the dataset page: https://huggingface.co/datasets/The-Adimension/prosat-nemotron-alice-in-wonderland-traces.nemotron-terminal-scientific_computing
nemotron-terminal-scientific_computing
Per-source partition of nvidia/Nemotron-Terminal-Corpus,
filtered to source == "scientific_computing". The difficulty column preserves the original
easy / medium / mixed split (na for the dataset_adapters/* files, which
did not carry a difficulty label).
Partitioning scheme:
adapters_{code,math,swe} — rows from dataset_adapters/{code,math,swe}.parquet
{skill} (e.g. debugging, security, …) — rows from
synthetic_tasks/skill_based/{easy,medium… See the full description on the dataset page: https://huggingface.co/datasets/laion/nemotron-terminal-scientific_computing.nemotron-nano2-safety-distill-gptoss
Nemotron Nano 2 Safety Distill — GPT-OSS
A distilled safety dataset produced using the Nemotron Nano 2 recipe with GPT-OSS-20B and GPT-OSS-120B as teacher models.
⚠️ Content Warning: This dataset includes potentially harmful prompts. Use responsibly for research purposes only.
Overview
This safety-focused distilled dataset was created by following the Nemotron Nano 2 safety recipe, adapted to use GPT-OSS-20B and GPT-OSS-120B as teacher models. Due to resource limitations… See the full description on the dataset page: https://huggingface.co/datasets/Ericwang/nemotron-nano2-safety-distill-gptoss.Nemotron-Terminal-Corpus2
Terminal-Corpus: Large-Scale SFT Dataset for Terminal Agents
Terminal-Corpus is a large-scale Supervised Fine-Tuning (SFT) dataset designed to scale the terminal interaction capabilities of Large Language Models (LLMs). Developed by NVIDIA, this dataset was built using the Terminal-Task-Gen pipeline, which combines dataset adaptation with synthetic task generation across diverse domains.
🚀 Key Results & Performance
The high-quality trajectories in Terminal-Corpus enable… See the full description on the dataset page: https://huggingface.co/datasets/CathleenTico/Nemotron-Terminal-Corpus2.nemotron-terminal-system_administration
nemotron-terminal-system_administration
Per-source partition of nvidia/Nemotron-Terminal-Corpus,
filtered to source == "system_administration". The difficulty column preserves the original
easy / medium / mixed split (na for the dataset_adapters/* files, which
did not carry a difficulty label).
Partitioning scheme:
adapters_{code,math,swe} — rows from dataset_adapters/{code,math,swe}.parquet
{skill} (e.g. debugging, security, …) — rows from
synthetic_tasks/skill_based/{easy,medium… See the full description on the dataset page: https://huggingface.co/datasets/laion/nemotron-terminal-system_administration.nemotron-terminal-software_engineering
nemotron-terminal-software_engineering
Per-source partition of nvidia/Nemotron-Terminal-Corpus,
filtered to source == "software_engineering". The difficulty column preserves the original
easy / medium / mixed split (na for the dataset_adapters/* files, which
did not carry a difficulty label).
Partitioning scheme:
adapters_{code,math,swe} — rows from dataset_adapters/{code,math,swe}.parquet
{skill} (e.g. debugging, security, …) — rows from
synthetic_tasks/skill_based/{easy,medium… See the full description on the dataset page: https://huggingface.co/datasets/laion/nemotron-terminal-software_engineering.ruler-100-nemotron
RULER-100 — Nemotron-Nano-v3 tokenized
RULER long-context evaluation data, regenerated with the
nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 (instruct) tokenizer so the labeled context
lengths are exact for that model — instead of drifting, as they do when RULER data tokenized for a
different model (e.g. Qwen3) is fed to Nemotron.
What's here
7 context lengths: 4096, 8192, 16384, 32768, 65536, 131072, 262144 (the model's max).
13 RULER tasks: niah_single_1/2/3… See the full description on the dataset page: https://huggingface.co/datasets/jet-ai/ruler-100-nemotron.nemotron-terminal-data_science
nemotron-terminal-data_science
Per-source partition of nvidia/Nemotron-Terminal-Corpus,
filtered to source == "data_science". The difficulty column preserves the original
easy / medium / mixed split (na for the dataset_adapters/* files, which
did not carry a difficulty label).
Partitioning scheme:
adapters_{code,math,swe} — rows from dataset_adapters/{code,math,swe}.parquet
{skill} (e.g. debugging, security, …) — rows from
synthetic_tasks/skill_based/{easy,medium… See the full description on the dataset page: https://huggingface.co/datasets/laion/nemotron-terminal-data_science.persona_nemotron
Persona Nemotron PT Datasets
This is a collection of Portuguese synthetic datasets, consisting of 3 datasets, one with general questions from varied topics, one with math questions, and one with instruction-following requests.
The prompts were generated using an approach similar to PersonaHub, with a translated version of Nemotron Personas. Both prompts and answers were generated using Gemma 3-27B.
This dataset is provided as part of the AMALIA project and is… See the full description on the dataset page: https://huggingface.co/datasets/amalia-llm/persona_nemotron.nvidia-Nemotron-Science-Math
NVIDIA Nemotron Science and Math Reasoning
This is an unofficial, curated collection derived from NVIDIA's open-source Nemotron datasets. It is designed specifically to train language models in complex scientific and mathematical reasoning by providing structured chain-of-thought (CoT) examples.
To ensure efficiency, the shortest available CoT sequence was chosen for each question, filtering out redundant variations while preserving the core logical progression toward the final… See the full description on the dataset page: https://huggingface.co/datasets/agentlans/nvidia-Nemotron-Science-Math.Nemotron-Terminal-Corpus
Terminal-Corpus: Large-Scale SFT Dataset for Terminal Agents
Terminal-Corpus is a large-scale Supervised Fine-Tuning (SFT) dataset designed to scale the terminal interaction capabilities of Large Language Models (LLMs). Developed by NVIDIA, this dataset was built using the Terminal-Task-Gen pipeline, which combines dataset adaptation with synthetic task generation across diverse domains.
🚀 Key Results & Performance
The high-quality trajectories in Terminal-Corpus enable… See the full description on the dataset page: https://huggingface.co/datasets/CathleenTico/Nemotron-Terminal-Corpus.nemotron-terminal-adapters_code
nemotron-terminal-adapters_code
Per-source partition of nvidia/Nemotron-Terminal-Corpus,
filtered to source == "adapters_code". The difficulty column preserves the original
easy / medium / mixed split (na for the dataset_adapters/* files, which
did not carry a difficulty label).
Partitioning scheme:
adapters_{code,math,swe} — rows from dataset_adapters/{code,math,swe}.parquet
{skill} (e.g. debugging, security, …) — rows from
synthetic_tasks/skill_based/{easy,medium… See the full description on the dataset page: https://huggingface.co/datasets/laion/nemotron-terminal-adapters_code.