This is a major update to Cog that removes the server/client model and makes it merely a wrapper around Docker. Kind of like Docker Compose for machine learning.
If you've been using the Cog server with a Docker registry, you now simply push directly to the registry by setting the image name in `cog.yaml`.
If you've got some models defined with Cog, here's what you need to update:
- `environment:` in `cog.yaml` is now `build:`
- `model:` in `cog.yaml` is now `predict:`
- The Python class `cog.Model` is now `cog.Predictor`.
Some things have moved around in the user interface too. Check out the docs to see how it works.
Changelog