Collaborative Domain Knowledge Tutor (Qwen2.5-7B-Instruct)

A fine-tuned Qwen2.5-7B-Instruct model specialized in collaborative problem-solving and domain knowledge tutoring, trained on 1,781 synthetic examples implementing the Usefulness, Inspiration, and Trust framework.

🎯 Model Description

This model has been fine-tuned to excel in collaborative learning scenarios, providing:

Step-by-step guidance for complex problems
Critical thinking stimulation and engagement
Real-world application of academic concepts
Collaborative planning and conflict resolution
Motivational support for learners

🏗️ Architecture

Base Model: Qwen/Qwen2.5-7B-Instruct
Training Method: LoRA (Low-Rank Adaptation)
Training Data: 1,781 synthetic collaborative learning examples
Framework: 11-aspect evaluation (Usefulness, Inspiration, Trust)

📊 Performance Improvements

Metric	Base Model	Fine-tuned	Improvement
Usefulness	0.147	0.400	+173.0%
Inspiration	0.125	0.144	+14.8%
Trust	0.067	0.043	-35.3%*
Overall Satisfaction	0.126	0.224	+78.1%
Collaborative Learning Score	0.108	0.180	+66.7%

*Note: Trust decrease is a known limitation being addressed in future iterations.

🚀 Usage

Load and Use the Model

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

# Load base model
base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-7B-Instruct")
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-7B-Instruct")

# Load LoRA adapters
model = PeftModel.from_pretrained(base_model, "christinekwon/collaborative-tutor-qwen25")

# Generate response
prompt = "Student: I'm struggling with this algebra problem. Can you help me work through it step by step?
Tutor:"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=300)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)

Example Prompts

Academic Support: "I don't understand how photosynthesis works. Can you explain it?"
Collaborative Planning: "We need to work together on this group project. How should we approach this?"
Creative Problem Solving: "We're stuck on this creative project. How can we think outside the box?"
Motivational Support: "I'm feeling discouraged about my progress. Can you help me stay motivated?"

🔬 Training Details

Training Data: 1,781 synthetic dialogue examples
Data Generation: Gemini 2.5 Pro API
Training Framework: PyTorch + Transformers
Optimization: LoRA with 4-bit quantization
Training Time: ~2 hours on Colab A100 GPU

📚 Research Context

This model implements the collaborative learning framework from our research on:

11 User-Perceived Aspects of educational interactions
Usefulness, Inspiration, and Trust as primary outcomes
Data-driven weights learned via Random Forest (R²=0.841)
Critical thinking as the highest-weighted aspect (0.157)

⚠️ Limitations

Trust scores show some decrease compared to base model
Peer teaching capabilities need improvement
Consistency across different domains varies
Memory requirements for full model loading

🔮 Future Work

Address trust and consistency issues
Expand training data with more diverse scenarios
Improve peer-to-peer learning capabilities
Optimize for lower memory requirements

📄 Citation

If you use this model in your research, please cite:

@misc{collaborative-tutor-qwen25,
  title={Collaborative Domain Knowledge Tutor: Fine-tuned Qwen2.5-7B-Instruct for Collaborative Learning},
  year={2025},
  url={https://huggingface.co/christinekwon/collaborative-tutor-qwen25}
}

📞 Contact

For questions, feedback, or collaboration opportunities, please reach out through the Hugging Face Hub.

This model is designed for educational and research purposes. Please use responsibly and in accordance with applicable guidelines.