- gpt4all-api
- RESTful API
- gpt4all-backend
- C/C++ (ggml) model backends
- gpt4all-bindings
- Language bindings for model backends
- gpt4all-chat
- Chat GUI
- gpt4all-docker
- Dockerfile recipes for various gpt4all builds
- gpt4all-training
- Model training/inference/eval code
This is roughly based on what's feasible now and path of least resistance.
-
Clean up gpt4all-training.
- Remove deprecated/unneeded files
- Organize into separate training, inference, eval, etc. directories
-
Clean up gpt4all-chat so it roughly has same structures as above
- Separate into gpt4all-chat and gpt4all-backends
- Separate model backends into separate subdirectories (e.g. llama, gptj)
-
Develop Python bindings (high priority and in-flight)
- Release Python binding as PyPi package
- Reimplement Nomic GPT4All to call new Python bindings
-
Develop Dockerfiles for different combinations of model backends and bindings
- Dockerfile for just model backend
- Dockerfile for model backend and Python bindings
-
Develop RESTful API / FastAPI