Suzy template design - PowerPoint Presentation

[https://github.com/alientony/SuzyAI/blob/main/Suzy\_template\_explanation.odp](https://github.com/alientony/SuzyAI/blob/main/Suzy_template_explanation.odp) [Suzy chat bot box](https://preview.redd.it/vc8s93fxbbmd1.png?width=449&format=png&auto=webp&s=0cc31340636ecc31ff65632132329eb7f1bd4321) Currently working on a template design for a specific use case of a chatbot. I'm creating a easily explained PowerPoint of points and design directions I've reasoned out for conversion strategies to make 7B / 8B models into specific use case machines. [HTML and Json formating](https://preview.redd.it/rl5moxu4dbmd1.png?width=1375&format=png&auto=webp&s=b8e95bd84eb65d5f6d9443696724406785d6e29d) I go into short explanations of what each part does and how the current format allows for certain features that can emerge from connection between the format and preexisting LLM structure. I was able to achieve a small working prototype from utilizing Unsloth's colab finetuning tools but without more data the results were not generalized enough to be deployed. This was using the LLAMA 3.1 8b model. Although after converting the model to a gguf it was substantially unusable due to nonsensical responses. [https://github.com/unslothai/unsloth](https://github.com/unslothai/unsloth) I haven't finished the Powerpoint yet but if there are any suggestions or comments please feel free to ask. Do you think it would be possible to delicate a single consumer grade model to a application? Is it necessary that the models be generally trained? Can a limited interface cause a especially skilled model or a overly fit tool?