---
license: mit
language: en
pipeline_tag: text-generation
base_model: unsloth/Qwen2.5-3B-Instruct
---


# Commgpt‑3B

**Commgpt‑3B** is a conversational language model fine-tuned from [unsloth/Qwen2.5-3B-Instruct](https://huggingface.co/unsloth/Qwen2.5-3B-Instruct), adapted specifically for **Advanced Communication Systems (EECE 442)** at the American University of Beirut.

The model was trained using a curriculum of domain-specific Q&A pairs and evaluated on a custom benchmark of 450 communication systems questions. It was originally deployed with a retrieval-augmented generation (RAG) pipeline and Gradio interface.

This repo contains **only the model weights** and a **detailed implementation report**.

📄 Full report and documentation:  
[https://github.com/DavidA00/Commgpt-final-year-project](https://github.com/DavidA00/Commgpt-final-year-project)

This work was part of a final year project conducted during Sept 2024 - May 2025. 

## Usage
```python
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline

model = AutoModelForCausalLM.from_pretrained("dabboud/Commgpt-3B")
tokenizer = AutoTokenizer.from_pretrained("dabboud/Commgpt-3B")

generator = pipeline("text-generation", model=model, tokenizer=tokenizer)
output = generator("What is Nyquist rate?", max_new_tokens=100)
print(output[0]["generated_text"])
```
---
## Citation
```latex
@misc{CommGPT2025,
  title   = {CommGPT: A Domain-Tuned Qwen 2.5-3B Model for Advanced Communication Systems},
  author  = {Abboud, David and Eid, Alex and Menassa, Alexander and Abou Faycal, Ibrahim and Fahs, Jihad and Zaraket, Fadi and Chokr, Sally},
  note    = {Model and report available at \url{https://huggingface.co/dabboud/Commgpt-3B}},
  year    = {2025}
}
```