OpenWeight

OpenWeight is the public release of a trained AI model’s final parameters (weights), allowing developers to download, fine-tune, and deploy the model independently without full access to the original training code.

OpenWeight defines the practice of releasing an AI model’s trained parameters, or 'weights,' which encode the model's learned knowledge across billions of variables. This approach is distinct from full open-source: it provides the final model brain for immediate use, fine-tuning, and security auditing, lowering the barrier to entry for developers. Key models like Meta's LLaMA 2 and Mistral 7B are distributed this way, granting the freedom to use and share the model. However, creators typically withhold the proprietary training data and complete training code, protecting their competitive process while still fostering community innovation.

https://opensource.org/blog/open-weights-not-quite-what-youve-been-told/