Nemotron Sft Multilingual V2: Translated Text for Fine-Tuning Language Models

A multilingual supervised fine-tuning dataset for text-generation models. It was generated by NVIDIA by translating seed data from three other Nemotron datasets to add coverage for Hindi, Korean, Brazilian Portuguese, and Japanese. The dataset was last updated on June 4, 2026.

Use Cases

• Fine-tuning language models for multilingual text generation based on translated seed data.
• Improving model performance in Hindi, Korean, Brazilian Portuguese, and Japanese language tasks.
• Post-training text-generation models using a dataset generated via a new data processing pipeline.

Strengths

• Dataset was created by NVIDIA, a major AI research organization.
• Covers multiple languages including Hindi, Korean, Brazilian Portuguese, and Japanese.
• Uses a new data processing pipeline that avoids line-breaking issues.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

NVIDIA

Collection Method

Generated by translating seed data from Nemotron-Math-v2, Nemotron-Competitive-Programming-v1, and Nemotron-Science-v1.

Freshness

Last updated 2026-06-04 04:47:54; freshness should be verified.