aiXcoder-7B is an open-source code large language model designed for code completion, comprehension, and generation across many programming languages. Released by the aiXcoder team, the model’s official repository on GitHub has around 2,300 stars and provides the weights, inference code, and fine-tuning examples needed to run or adapt it locally.
Unlike general chat models later pointed at code, aiXcoder-7B was trained specifically for software development tasks, with an emphasis on practical completion quality โ including method-level and file-level completion that considers surrounding context. The same model powers aiXcoder’s IDE plugins for VS Code and JetBrains editors, and the open weights let teams self-host it where code cannot leave their network.
The model was evaluated by its creators on standard code benchmarks covering completion and generation, and the repository documents both GPU inference and quantized deployment through bitsandbytes for smaller machines. aiXcoder also publishes the data recipe behind the model, which is useful for teams that want to replicate a code-LLM training run or continue training the model on an internal codebase with the provided fine-tuning examples.
The repository includes environment setup instructions and a Docker option, and the model follows the conventions of the aiXcoder product line, which has been used for code completion in production IDE settings. For researchers, the release is notable for documenting how a relatively small 7B model, trained with completion-specific data mixes, can rival much larger general models on completion tasks.
aiXcoder-7B is free and open source under the Apache-2.0 license, including the downloadable model weights on Hugging Face.
Developers and platform teams who want a self-hostable code completion model, or researchers studying code LLMs and repository-level completion.