ServiceNow, Hugging Face, and NVIDIA release new open LLMs for developers


ServiceNow, Hugging Face, and NVIDIA have teamed up to release a new family of open LLMs called StarCoder2 that is designed for developers. 

StarCoder2 was trained on 619 programming and is intended to provide developers with features like code generation, workflow generation, and text summarization, to name a few. The companies envision the StarCoder2 models will be useful to both software engineers and citizen developers. 

It was developed within the BigCode community, which is a group devoted to responsibly developing LLMs. The project was stewarded by both ServiceNow and Hugging Face. 

StarCoder 2 comes in three different model sizes: ServiceNow trained a 3 billion-parameter model, Hugging Face trained a 7 billion-parameter model, and NVIDIA trained a 15 billion-parameter model.

The smaller models are designed to offer powerful performance while using small amounts of compute power. According to the companies, the 3 billion-parameter model matches the performance of the 15 billion-parameter model of the original StarCoder release

Users will be able to fine-tune these models to meet their own specific needs, using open-source tools such as NVIDIA NeMo or Hugging Face TRL. 

“StarCoder2 stands as a testament to the combined power of open scientific collaboration and responsible AI practices with an ethical data supply chain,” said Harm de Vries, lead of ServiceNow’s StarCoder2 development team, and co-lead of BigCode. “The state-of-the-art open-access model improves on prior generative AI performance to increase developer productivity and provides developers equal access to the benefits of code generation AI, which in turn enables organizations of any size to more easily meet their full business potential.”

Leandro von Werra, machine learning engineer at Hugging Face and co‑lead of BigCode, added: “The joint efforts led by Hugging Face, ServiceNow and NVIDIA enable the release of powerful base models that empower the community to build a wide range of applications more efficiently with full data and training transparency. StarCoder2 is a testament to the potential of open‑source and open science as we work toward democratizing responsible AI.”



Source link