This is a LoRA for the Llama-2-70b-Guanaco LLM. This is not a LoRA for Stable Diffusion.
This is a Medical Knowledge LoRA made for use with this model: llama-2-70b-Guanaco-QLoRA-fp16 https://huggingface.co/TheBloke/llama-2-70b-Guanaco-QLoRA-fp16) (you can use the quantized version of the model too).
I'm experimenting with making LLM LoRAs and thought this one tuned out good enough to share. Use at your own risk however, I am not a medical expert and do not take responsibility for the model producing incorrect information.
This LoRA was made using Oobabooga (https://github.com/oobabooga/text-generation-webui) and these training data: https://huggingface.co/datasets/BI55/MedText
I'll be working on various versions of this LoRA (and other types of LoRAs) as I explore the various ways in which to construct these add-on models.
The Hugging Face page for all this is here:
https://huggingface.co/AARon99/MedText-llama-2-70b-Guanaco-QLoRA-fp16
If you too are curious on how to construct these LLM based LoRAs, the Hugging Face page has additional information intended to help beginners, like myself.
Description
This LoRA was trained using the "Raw text file" option in Oobabooga, I'm currently working on a LoRA that uses "Formatted Data" but this process takes a lot longer.
Simply load the LoRA with your model and start asking it medical questions in the "instruct" conversation field of Oobabooga and watch as you get pretty decent responses, best inference parameters are those from "Divine Intellect"
FAQ
Comments (5)
It's unexpected to see this here. I'm interested in seeing more LLM loras.
Thank you! I was hesitant at first, but figured I'd upload it anyway. Oh boy do I have some interesting things in the works! I'm working on a LoRA that will generate good stable diffusion inputs for image generation and other types of LoRAs.
The image generation one I want to work well with the Auto1111 extension for Oobabooga!
厉害，有的人来这只是为了色情（比如我），而你，我的朋友，你是真正的英雄
@wyxzddsjj919 :3 thank you! I like experimenting with image and language models. Civitai said they were adding LLM stuff to the site soon, which I think is a great idea.
I don't think I can run a 70b local but still very cool. Thank you.