Summary

StarChat is a series of language models from Hugging Face and BigCode that act as helpful coding assistants, with the original playground demo built on StarCoder.

Key Features

Coding assistant language model fine-tuned from StarCoder / StarCoder2
16B-parameter model pre-trained on one trillion tokens of code and text
Support for 80+ programming languages in the original version, 600+ in StarChat2
Code generation and completion from natural-language prompts
Conversational chat about programming topics
Fine-tuned on Dolly and OpenAssistant dialogue datasets
Open model under the BigCode Open RAIL-M licence
Interactive playground demo on Hugging Face Spaces
Educational tool for learning about LLM fine-tuning

About StarChat


The alpha version is a 16-billion-parameter variant of StarCoderBase, pre-trained on one trillion tokens from over 80 programming languages, GitHub issues, commits and Jupyter notebooks, and fine-tuned on the Dolly and OpenAssistant datasets to act as a coding assistant. Later versions, including StarChat2, are fine-tuned from StarCoder2 and cover over 600 programming languages.


StarChat is provided as an educational tool for explaining large language model fine-tuning, not as a replacement for human expertise, and the alpha version was not aligned with RLHF, so it can produce problematic or semantically incorrect code.

Get more likes & reach the top of search results by adding this button on your site!

Embed button preview - Light theme
Embed button preview - Dark theme
TurboType Banner

Subscribe to the AI Search Newsletter

Get top updates in AI to your inbox every weekend. It's free!