This modelfile is the LoRA adapter for Andy-3.5-reasoning
Why this exists
This Repo exists because I wanted to make Andy-3.5, as well as its derivitives, such as Andy-3.5-reasoning, fully open-source. Via Unsloth, you are able to continue fine tuning where I left off, so if you made your own dataset, you can continue tuning Andy-3.5 for your exact use case.
What if I fine tune off of Andy-3.5?
If you fine tune Andy-3.5 on your dataset, my dataset, or any other dataset, you have to provide credit to me for making the base model, which is Andy-3.5, if you wish, you may call the model Andy-3.5-base
Why would I want to fine tune off of Andy-3.5?
Andy-3.5 has a significant amount of knowledge regarding Minecraft and MindCraft, but not unlimited. Andy-3.5 can be trained further on Minecraft knowledge to make the model better, and if you strive for maximum efficiency, it would be best to continue fine-tuning a model based on similar data to help it.
What should I call my model if I do tune it?
You may name it whatever you'd like, but if I may suggest, I would recommend a name that clearly references the fact it originated from Andy-3.5.
If you'd like an example, if I trained Andy-3.5 on speedrunning tactics, I would call the model Andy-3.5-Speedrun or something similar.
Important notes:
- I do not suggest fine tuning off of this model for anything besides reasoning
- I do not suggest fine tuning this model with any dataset for reasoning that does not use the DeepSeek-R1 method of thinking.