ServiceNow, Hugging Face, and NVIDIA have unveiled StarCoder2, a large language model (LLM) available for open-access use, with a focus on code writing tasks. The standout features include operational efficiency, AI transparency, and cost management assistance.
StarCoder2, a project by the BigCode community supported by ServiceNow and Hugging Face, has been trained on programming languages in 619 languages. It is designed to work seamlessly with applications within organizations for various tasks such as code generation, workflow creation, content summarization, and more. Organizations can leverage it with moderate resource customization.
StarCoder2 offers models in 3 sizes. The first model consists of 3 billion parameters trained with ServiceNow’s Fast LLM framework, the 7 billion parameters model trained with Hugging Face’s nanotron framework, and the largest 15 billion parameters model created from NVIDIA NeMo and trained on NVIDIA’s cloud resources.
StarCoder2’s dataset has been enhanced from previous versions to understand low-resource languages like COBOL and perform mathematical functions.
Available under the BigCode Open RAIL-M license without any usage fees, StarCoder2 can be accessed for free. For more information, visit the project’s GitHub page.
Source: ServiceNow
TLDR: ServiceNow, Hugging Face, and NVIDIA introduce StarCoder2, a large language model for code writing tasks, available for free, with a focus on efficiency, AI transparency, and cost management.
Leave a Comment