AiAF/First-Finetune_QLoRA_4bit

I've always wanted to take a crack at finetuning an LLM So this is my first successful one (sort of). This is a Qlora fine tune of Mistral-7B-v0.1 Trained on information about alien abductions. Why aliens? And wanted to pick something relatively niche. something that I knew the. model wouldn't have A ton of information on. If I picked something more well-known or mainstream like Bitcoin or something like that, then it would be kind of hard to evaluate whether or not my fine tuning actually did anything because the model almost certainly knows a lot of stuff about Bitcoin already. It's obviously not perfect, and I only used one singular data set because it took me forever to even figure out how this stuff works in the 1st place. I will test more tommorow To see if longer training results and better results. I don't think it's a subject matter expert on the documents I trained it on. but it is certainly able to recall information. from the documents come up like for example what an "Enforcer Hybrid" is, So I know it at least worked to some extent. The data set was generated using Heralax's (https://huggingface.co/Heralax) Augmentoolkit (https://github.com/e-p-armstrong/augmentoolkit). I'll add any other relevant information later on, but I think this is pretty much it.