SUSE has announced the release of the large language model Cavil-Qwen3-4B under the Apache 2.0 license, used in SUSE and openSUSE projects within the Cavil toolkit for analyzing license compliance in code. The published model encompasses 4 billion parameters and is based on the Qwen3-4B model, further optimized for text classification.
The primary purpose of the model is to identify licenses used in the source code of software and documentation. To accomplish this task, the model has been additionally trained on a dataset containing 150,000 examples of headings and comments referencing licenses in source code. In practice, the model automates the verification of license compliance in the codebase to identify license incompatibilities and potential legal issues with the code.
The model's size is optimized to achieve a balance between high-quality understanding of language constructs and the ability to run on typical consumer GPUs. In addition to the model itself, a publicly available dataset used in training is provided, along with tools for validation. A handler is also available to leverage the model within the Cavil toolkit, designed for checking source code for compliance with legal norms and requirements (license checks, identifying license violations, risk assessment).
Source: opennet.ru
