Suggested Learning Order:
-
01-openai: all notebooks. This is the OpenAI GPT API.
-
03-llama-cpp: all notebooks and scripts. Llama.cpp is significant in that it's the first project that made CPU inference possible. You can run these if you have a laptop/desktop with >= 16GB memory. The Python library is largely compatible to OpenAI and has a grammar API that can be very handy.
-
05-transformers: all notebooks and scripts. This is the primary API researchers use. All new techniques are first implemented here. For local training or finetune this is the only choice.
-
06-vLLM: all notebooks. vLLM is the goto framework for local model deployment. It is very easy to use and very efficient. If you developed your method with transformers and then port to vLLM you might see change of accuracy.
Google's Gemini and Gemma can be safely skipped.