The Nemotron-Pretraining-Code-v3 dataset is part of the Nemotron Pretraining Data collection of pretraining datasets. Designed for the NVIDIA Nemotron 3 family of LLMs, this dataset is intended to improve the coding capabilities of LLMs.
The Nemotron-Pretraining-Code-v3 dataset contains the metadata corresponding to the raw source-code update to our Nemotron-Pretraining-Code-v2 and Nemotron-Pretraining-Code-v1… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-Pretraining-Code-v3.