Search code, repositories, users, issues, pull requests...
The ollama registry is somewhat a docker registry, but also somewhat not. Hence, normal docker pull through caches do not work. This project aims to pull through cache, that you can deploy in your local network, to speed up the pull times of the ollama registry.
p.S. Please give a thumbs up on this PR, so that the default behavior of the ollama client can be overwrite to use the cache. Will look nicer and work better.
This proxy is based on a worker architecture. It has a worker pool, that can be configured via the NUM_DOWNLOAD_WORKERS environment variable. When a request comes in, the proxy will check if the file is already in the cache. If not, it will mark it as queued and serve the request from the upstream. In the background the worker checks the queue and downloads the file. Going forward, the file will be served from the local cache, not upstream anymore
