HuggingFace and ServiceNow reveal Github Copilot rival
GitHub Copilot finally has a worthy rival thanks to a collaboration between Hugging Face and ServiceNow, who formed a community of developers that go by the name BigCode.
Despite having no affiliation with GitHub, the StarCoder and StarCoderBase code LLMs were trained on data from GitHub, which the team says was “permissively licensed,” likely in a nod to the legal action that GitHub faced for failing to attribute some of the code it used.
More than 80 programming languages, Git commits, GitHub issues, and Jupyter notebooks went into creating the AI assistant for developers over the course of more than six months, with a separate version trained specifically with Python in mind.
StarCoder takes on GitHub Copilot
According to the announcement (opens in new tab), StarCoder was found to have outperformed other existing open code LLMs in some cases, including the OpenAI model that powered early versions of GitHub Copilot.
It also promises to be able to process more code than other tools, thanks to a context length of over 8,000 tokens.
As well as being able to write, suggest, and fix code, Hugging Face also explained that StarCoder can act as a tech assistant, answering programming-related questions to help in other areas of developers’ workflows.
An early v0.1 version of HuggingChat can be tried online (opens in new tab) which opens using ‘OpenAssistant/oasst-sft-6-llama-30b’ that promises to be “a good alternative to ChatGPT.” There is an option to change the model to ‘BigCode/StarCoder’, the self-proclaimed “good model for answering technical questions.”
BigCode (opens in new tab), the body behind the model, is a project intended to responsibly develop LLMs led by ServiceNow and Hugging Face. Besides the core members, it invites contributors and AI researchers to collaborate openly.