Datasetpublic

Arsh9210/Nemotron-RL-Instruction-Following-Structured-Outputs-v2

Dataset Description: Split 1: Direct Generation tests the model’s ability to perform freeform text structured outputs on JSON, YAML, and XML data, varying the complexity and presentation of the schema. Split 2: Diversified Tasks adds 2 additional output formats: TOML and CSV, while increasing problem types to Direct Extraction from document, Translation between formats, Multistep Translation from known data, Multistep Extraction from unrelated context, Schema-Only Generation for… See the full description on the dataset page: https://huggingface.co/datasets/Arsh9210/Nemotron-RL-Instruction-Following-Structured-Outputs-v2.

sourceHugging Facecc-by-4.0updated 2mo agoView on Hugging Face

0likes27downloads

Dataset Card

Dataset Description:

Split 1: Direct Generation tests the model’s ability to perform freeform text structured outputs on JSON, YAML, and XML data, varying the complexity and presentation of the schema.

Split 2: Diversified Tasks adds 2 additional output formats: TOML and CSV, while increasing problem types to Direct Extraction from document, Translation between formats, Multistep Translation from known data, Multistep Extraction from unrelated context, Schema-Only Generation for realistic-looking data generation, and Error Correction from corrupted output to match a given schema.

Split 3: Tool-Calling Extraction tasks the model with document summary and extraction using a dedicated extraction tool, testing the model's ability to correctly match complex and deep schemas for tool calling, with and without distractors.

This dataset is ready for commercial or non-commercial uses.

Dataset Owner(s):

NVIDIA Corporation

Dataset Creation Date:

Created on: April 15, 2026 Last Modified on: April 27, 2026

Version:

Nemotron-RL-Instruction-Following-Structured-Outputs-v2

Previous Version(s): nvidia/Nemotron-RL-instruction_following-structured_outputs

This dataset is a direct successor to nvidia/Nemotron-RL-instructionfollowing-structuredoutputs and may be used as a replacement or supplement to it.

License/Terms of Use:

Governing terms: this dataset is licensed under CC BY 4.0.

Intended Usage:

Reinforcement learning training for instruction following capabilities, especially in structured outputs with diverse output types, and structured outputs with tool-use extraction tasks.

Dataset Characterization

Data Collection Method Synthetic

Labeling Method Hybrid: Synthetic, Automatic

Dataset Format

Modality: Text Format: JSONL Structure: Text + Metadata

Dataset Quantification

1.Split 1: Direct Generation

Subset	Samples	Size
JSON output	9,382 (33.3%)	91.53 MB (0.092 GB)
XML output	9,382 (33.3%)	94.10 MB (0.094 GB)
YAML output	9,381 (33.3%)	92.00 MB (0.092 GB)
Total	28,145	277.63 MB (0.278 GB)

1.Split 2: Diversified Tasks

Subset	Samples	Size
JSON output	6,734 (26.1%)	93.64 MB (0.094 GB)
YAML output	5,525 (21.4%)	83.06 MB (0.083 GB)
XML output	5,173 (20.1%)	82.35 MB (0.082 GB)
TOML output	4,177 (16.2%)	69.73 MB (0.070 GB)
CSV output	4,159 (16.1%)	69.29 MB (0.069 GB)
Total	25,768	398.07 MB (0.398 GB)

1.Split 3: Tool-Calling Extraction

Subset	Samples	Size
Random wrapper tool schema	3,478 (39.6%)	108.72 MB (0.109 GB)
Multi-key object tool schema	2,600 (29.6%)	74.97 MB (0.075 GB)
Extraction wrapper tool schema	2,162 (24.6%)	67.97 MB (0.068 GB)
Direct tool schema	543 (6.2%)	17.06 MB (0.017 GB)
Total	8,783	268.72 MB (0.269 GB)

Reference(s):

Nemo-Gym configs:

1.Split 1: Direct Generation:
2.https://github.com/NVIDIA-NeMo/Gym/blob/main/resourcesservers/structuredoutputs/configs/structuredoutputsjsonyamlxml_v1.yaml
3.Split 2: Diversified Tasks
4.https://github.com/NVIDIA-NeMo/Gym/blob/main/resourcesservers/structuredoutputs/configs/structuredoutputsv3.yaml
5.Split 3: Tool-Calling Extraction
6.https://github.com/NVIDIA-NeMo/Gym/blob/main/resourcesservers/structuredoutputs/configs/structuredoutputsv4.yaml

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. 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 model quality, risk, security vulnerabilities or NVIDIA AI Concerns here.