Generate predict.py and cog.yaml automatically using GPT4

pip install autocog

First, set your OpenAI API key in an environment variable

OPENAI_API_KEY=sk-...

In the repo you want to cog-ify, run

autocog

This will generate a cog.yaml and predict.py based on the files in the current directory. It will then run the model and if it fails to run, it will attempt to fix the error and run it again. By default it has 5 attempts to fix it, which can be changed with the --attempts flag.

If your model needs a GPU to run, you need to run AutoCog on a GPU machine.

Sometimes AutoCog fails to create a working Cog configuration. In those cases you, the human, have to step in and edit the cog.yaml and predict.py files.

Once you have edited them, let AutoCog continue by running autocog again. If you'd like to recreate predict.py and cog.yaml from scratch, run autocog --initialize.

By default, AutoCog will guess a cog predict command to run the model. If you want to specify your own predict command, use the --predict-command flag.

If you want AutoCog to take the generation of cog.yaml and predict.py in a specific direction, you can use the --tell flag to prompt GPT4:

autocog --tell="Add inputs to allow for inpainting"