Nemotron Cascade 2 Supervised Fine-Tuning Data for Math and Science

This is the supervised fine-tuning data used for training the Nemotron-Cascade-2 model. It contains prompts and responses for math and science topics, with math responses generated by models including DeepSeek-V3.2 and GPT-OSS-120B.

Use Cases

• Fine-tune language models on math prompts with generated responses from DeepSeek-V3.2 and GPT-OSS-120B.
• Train models on science prompts collected from the described sources for domain-specific instruction following.
• Use the math proof prompts and responses from Nemotron-Math-Proofs-v1 for reasoning-focused model training.

Strengths

• Data is sourced from established prior datasets including Nemotron-Cascade-1-SFT and Nemotron-Math-v2.
• Math responses are generated by multiple high-capacity models such as DeepSeek-V3.2-Speciale.
• The dataset was last updated in March 2026.

Limitations

• The total number of rows, columns, and overall dataset size are unknown.
• Sample data and specific file formats are unavailable for inspection.
• The license and detailed data collection methodology are not specified.

Provenance

Source

nvidia

Collection Method

Prompts sourced from prior Nemotron datasets; responses generated by specified language models.

Freshness

2026-03-19