Nemotron-4 340B Technical Report
Abstract
We release the Nemotron-4 340B model family, including Nemotron-4-340B-Base, Nemotron-4-340B-Instruct, and Nemotron-4-340B-Reward. Our models are open access under the NVIDIA Open Model License Agreement, a permissive model license that allows distribution, modification, and use of the models and its outputs. These models perform competitively to open access models on a wide range of evaluation benchmarks, and were sized to fit on a single DGX H100 with 8 GPUs when deployed in FP8 precision. We believe that the community can benefit from these models in various research studies and commercial applications, especially for generating synthetic data to train smaller language models. Notably, over 98% of data used in our model alignment process is synthetically generated, showcasing the effectiveness of these models in generating synthetic data. To further support open research and facilitate model development, we are also open-sourcing the synthetic data generation pipeline used in our model alignment process.
Cite
@article{arxiv.2406.11704,
title = {Nemotron-4 340B Technical Report},
author = {Nvidia and : and Bo Adler and Niket Agarwal and Ashwath Aithal and Dong H. Anh and Pallab Bhattacharya and Annika Brundyn and Jared Casper and Bryan Catanzaro and Sharon Clay and Jonathan Cohen and Sirshak Das and Ayush Dattagupta and Olivier Delalleau and Leon Derczynski and Yi Dong and Daniel Egert and Ellie Evans and Aleksander Ficek and Denys Fridman and Shaona Ghosh and Boris Ginsburg and Igor Gitman and Tomasz Grzegorzek and Robert Hero and Jining Huang and Vibhu Jawa and Joseph Jennings and Aastha Jhunjhunwala and John Kamalu and Sadaf Khan and Oleksii Kuchaiev and Patrick LeGresley and Hui Li and Jiwei Liu and Zihan Liu and Eileen Long and Ameya Sunil Mahabaleshwarkar and Somshubra Majumdar and James Maki and Miguel Martinez and Maer Rodrigues de Melo and Ivan Moshkov and Deepak Narayanan and Sean Narenthiran and Jesus Navarro and Phong Nguyen and Osvald Nitski and Vahid Noroozi and Guruprasad Nutheti and Christopher Parisien and Jupinder Parmar and Mostofa Patwary and Krzysztof Pawelec and Wei Ping and Shrimai Prabhumoye and Rajarshi Roy and Trisha Saar and Vasanth Rao Naik Sabavat and Sanjeev Satheesh and Jane Polak Scowcroft and Jason Sewall and Pavel Shamis and Gerald Shen and Mohammad Shoeybi and Dave Sizer and Misha Smelyanskiy and Felipe Soares and Makesh Narsimhan Sreedhar and Dan Su and Sandeep Subramanian and Shengyang Sun and Shubham Toshniwal and Hao Wang and Zhilin Wang and Jiaxuan You and Jiaqi Zeng and Jimmy Zhang and Jing Zhang and Vivienne Zhang and Yian Zhang and Chen Zhu},
journal= {arXiv preprint arXiv:2406.11704},
year = {2024}
}