Welcome to our lil' ol' GitHub repository. Here, we're gonna be usin' them fancy GPT-4 models provided by the kind folks over at Nomic AI and them tokenizers from Hugging Face.
If y'all are itchin' to get them models, well, you're in luck! Just head on over to this here link:
https://the-eye.eu/public/AI/models/nomic-ai/gpt4all/
Now, we're gonna need us a tokenizer to get this party started. You can find that right here:
https://huggingface.co/decapoda-research/llama-7b-hf/tree/main
Y'all are gonna find some useful scripts in the /submodules/pyllamacpp/pyllamacpp/scripts
folder. They're labeled like this:
convert_gpt4all.py convert.py migrate.py
To convert the file to a usable model with LangChain, run this here command in the scripts folder while in a virtual Python environment:
python convert_gpt4all.py ../../../models/gpt4all-lora-unfiltered-quantized.bin ../../../submodules/llama-7b-hf/tokenizer.model ../../../models/gpt4all-lora-unfiltered-quantized-fuckedwith.bin
We got some handy-dandy IPython files for y'all to see how to get them llama_cpp and gpt4all workin' together. Check 'em out in this here repository.
Y'all have fun now, ya hear! 🤠