redmint/studybuddy-V2

StudyBuddy-AI v2 (LoRA Fine-Tuned Upgrade)

This model is an upgraded version of redmint/studybuddy-ai, fine-tuned using LoRA adapters to improve instruction following, truthfulness, and multi-step reasoning performance.

What This Version Improves

• Stronger multi-step reasoning using a 500-example custom reasoning dataset.
• Higher factual accuracy using a 500-example faithfulness dataset.
• Better general instruction handling using a 1000-example general dataset.
• More stable generation thanks to improved tokenization and padding setup.

Training Setup

• Base model: redmint/studybuddy-ai

• Method: LoRA fine-tuning

• Hardware: Google Colab T4

• Steps: ~500 for Stage 1 + ~500 for Stage 2

• Datasets:

  • general.jsonl (1000 lines)
  • multi-stepV2.jsonl (500 lines)
  • stayfaithful.jsonl (500 lines)

Files in This Model

• LoRA weight files (adapter_config.json, adapter_model.bin)
• Tokenizer files (tokenizer.json, special_tokens_map.json, etc.)
• Training configuration and HF metadata

🚀 How to Use

from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel

base = AutoModelForCausalLM.from_pretrained("redmint/studybuddy-ai")
peft = PeftModel.from_pretrained(base, "redmint/studybuddy-v2")

tokenizer = AutoTokenizer.from_pretrained("your-username/studybuddy-v2")

input_ids = tokenizer("Explain gravity.", return_tensors="pt").input_ids
print(tokenizer.decode(peft.generate(input_ids, max_new_tokens=200)[0]))

Notes

• This repo contains LoRA adapters, not a full standalone model.
• To deploy the upgraded model, always load the base model + these LoRA weights.

License

This model is licensed under the Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0).

You are free to: -Use the model for personal or educational purposes -Modify, remix, or build upon it

You may not:

• Sell, sublicense, or use the model or its outputs for any commercial purposes

© 2025 Micah Miles

NOTICE

This project is licensed for non-commercial use only.

For commercial licensing or partnership inquiries, contact: [email protected]

Message to future me, "I appreciate the disciplne and resilience you've managed to maintain during this whole project regardless of the trials and tribulations, and for that reason regardless of how it goes from here.. I'm proud of you dawg"😂 Keep shooting for the stars to anyone else that is reading this, and always remember.. SKY IS NOT THE LIMIT