VolkerMauel/nvidia-Nemotron-Agentic-v1-arrow

Arrow compatible version of nvidia/Nemotron-Agentic-v1.

from datasets import load_dataset
ds = load_dataset("VolkerMauel/nvidia-Nemotron-Agentic-v1-arrow", split="tool_calling")

should work fine on this one. 250MB Shards.

===

Dataset Description:

The Nemotron-Agentic-Tool-Use-v1 dataset is designed to strengthen models’ capabilities as interactive, tool-using agents. It focuses on multi-turn conversations where language models decompose user goals, decide when to call tools, and reason over tool outputs to complete tasks reliably and safely.

This dataset is ready for commercial use.

The Nemotron-Agentic-Tool-Use-v1 dataset contains the following subsets:

Interactive Agent

This dataset consists of synthetic multi-turn trajectories for conversational tool use, created by simulating three roles with large language models: a user given a task to accomplish, an agent instructed to help complete that task, and a tool execution environment that responds to the agent’s tool calls. Each trajectory captures the full interaction between these entities.

To ensure that the trajectories are high quality and that every action is consistent with each actor’s goals, we employ a separate language model as a judge to score and filter the data, removing trajectories where any step appears inconsistent, incoherent, or that use the incorrect tools. We use Qwen3-235B-A22B-Thinking-2507, Qwen3-32B, GPT-OSS-120B, and Qwen3-235B-A22B-Instruct-2507 both to generate the synthetic interactions and to support this judging process, resulting in a dataset that emphasizes reliable, goal-aligned conversational tool use.

Tool calling

The general-purpose tool-calling subset is generated using a similar method as the Interactive Agent subset. We collect tool sets from publicly available datasets and simulate conversations involving tool use. This subset uses Qwen3-235B-A22B-Thinking-2507 and Qwen3-235B-A22B-Instruct-2507 for simulating the conversation as well as turn-level judgements. The user simulator LLM is seeded with a user persona from nvidia/Nemotron-Personas-USA.

Dataset Owner(s):

NVIDIA Corporation

Dataset Creation Date:

Created on: Dec 3, 2025 Last Modified on: Dec 3, 2025

License/Terms of Use:

This dataset is governed by the Creative Commons Attribution 4.0 International License (CC BY 4.0). Additional Information: Apache 2.0 License for Glaiveai/Glaive-Function-Calling-v2.

Intended Usage:

This dataset is intended for LLM engineers and research teams developing and training models for agentic workflows and conversational tool use. It is suitable for supervised fine-tuning, data augmentation, and evaluation of models that must plan, call tools, and reason over multi-step interactions while staying aligned with the user and available tools in an environment. The trajectories can be used to train end-to-end tool-using assistants, build and benchmark tool-use planners or controllers, and study robustness of multi-role agent setups.

Dataset Characterization

Data Collection Method Hybrid: Human, Synthetic, Automated

Labeling Method Hybrid: Human, Synthetic, Automated

Dataset Format

Modality: Text Format: JSONL Structure: Text + Metadata

Dataset Quantification

Total Disk Size: ~ 5.5GB

Ethical Considerations:

NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their internal developer teams to ensure this dataset meets requirements for the relevant industry and use case and addresses unforeseen product misuse. Please report quality, risk, security vulnerabilities or NVIDIA AI Con