BigCode

BigCode is an open science collaboration dedicated to the responsible development of large-scale code language models, empowering the machine learning and open source communities through open governance.

Code LLM is able to complete and synthesize code from other code snippets and natural language descriptions, and works across a wide range of domains, tasks and programming languages. For example, these models can help professional and citizen developers build new applications.

BigCode invites AI researchers to co-develop state-of-the-art code LLMs and collaborate on the following research topics:

Build a representative evaluation suite for Code LLM, covering a diverse set of tasks and programming languages

Develop new methods to train and infer LLM faster

Legal, ethical and governance aspects of the LLM Code

The BigCode project is conducted in the spirit of open science. Datasets, models, and experiments are developed through open collaboration and released to the community under license. While the project has corporate support from ServiceNow and HuggingFace, such as hosting models and datasets, as well as training computations; all technical governance is conducted in working groups and working groups across the community.

Because Code LLM is developed with data from the open source community, we believe open governance can help ensure these models benefit the larger developer community. We are developing tools that give code creators the power to decide whether their source code is included in the training data and attribute model output to the developer when it is close to a copy of the training data.