From Nvidia:

ServiceNow, Hugging Face, and NVIDIA collaborate to release StarCoder2, a family of large language models for code generation. StarCoder2 is trained on 619 programming languages and offers three model sizes. The models can be further trained and used in enterprise applications to accelerate innovation and productivity.

StarCoder2 sets new standards for performance, transparency, and cost-effectiveness in code generation. The open-access model emphasizes responsible AI practices and ethical data supply chains. The collaboration between ServiceNow, Hugging Face, and NVIDIA aims to democratize responsible AI and empower developers with equal access to code generation AI benefits.

Developers can fine-tune the open-access StarCoder2 models with industry-specific data using open-source tools. Organizations have started using these models to create specialized capabilities for their businesses. ServiceNow and Hugging Face have already utilized the models to develop purpose-built solutions like text-to-code and chat assistant features.

BigCode, led by Hugging Face and ServiceNow, promotes open scientific collaboration in AI. The BigCode community contributed to the technical aspects of the StarCoder2 project and leveraged frameworks to train the models responsibly. StarCoder2 was built using responsibly sourced data from Software Heritage, aligning with policies for ethical AI development.

StarCoder2, like its predecessor, will be available under the BigCode Open RAIL-M license for royalty-free access. The model’s supporting code will be hosted on the BigCode project’s GitHub page for transparency and collaboration. All StarCoder2 models will also be available for download from Hugging Face and NVIDIA AI Foundation models for developers to experiment with.



Read more at Nvidia: ServiceNow, Hugging Face, and NVIDIA Release New Open-Access LLMs to Help Developers Tap Generative AI for Building Enterprise Applications