Nemotron-Pretraining-Code-v3: Metadata for LLM Coding Capability Training

A dataset from NVIDIA, last updated on 2026-06-04, designed to improve the coding capabilities of large language models. It contains metadata corresponding to a raw source-code update for the NVIDIA Nemotron 3 family of models. The dataset is part of the Nemotron Pretraining Data collection.

Use Cases

• Fine-tuning code-specialized LLMs based on the described source-code metadata
• Improving model performance on programming tasks using the updated pretraining data
• Benchmarking coding capability improvements across model versions

Strengths

• Designed specifically for the NVIDIA Nemotron 3 family of LLMs
• Represents an update to previous versions (v1 and v2), suggesting iterative improvement

Limitations

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

Provenance

Source

nvidia

Freshness

Last updated 2026-06-04 05:22:40