---
base_model: unsloth/qwen2.5-7b-unsloth-bnb-4bit
tags:
- text-generation-inference
- transformers
- unsloth
- qwen2
- gguf
license: apache-2.0
language:
- en
---

# Finance Earnings Call Q&A Bot

**Fine-tuned LLM for financial question-answering and earnings call simulation**  
By Mrinal Kumar

---

## 🚀 Model Overview

This model is a quantized [Qwen-2.5B / your base model] transformer fine-tuned on real earnings call transcripts and Q&A pairs from S&P 500 companies and global markets. It is designed to:

- Summarize complex financial calls into key insights
- Simulate Q&A between analysts and CFOs/CEOs

---

## 🏆 Example Use Cases

- **Students:** Learn how real-world analysts and CFOs communicate  
- **Investors:** Get concise summaries or simulate earnings call Q&A  
- **Researchers:** Build finance chatbots or extract structured knowledge from transcripts

---

## 🗂️ Training Data

- **Dataset:** Manually curated Q&A pairs extracted from publicly available earnings calls ([Kaggle Earnings Call Datasets](https://www.kaggle.com/))
- **Format:** Each example consists of an analyst question (`input`) and a CFO/CEO response (`output`)
- **Size:** 1,000+ Q&A pairs for diverse scenarios

---

## 💡 Example Questions

| Analyst Question                                     | Model Response                                                     |
| ---------------------------------------------------- | ------------------------------------------------------------------ |
| What drove the 20% YoY revenue growth?               | Revenue growth was driven by new subscriptions and higher pricing. |
| Can you explain margin contraction in the EU market? | Margins contracted due to logistics costs and currency headwinds.  |
| What are your key risks for next quarter?            | Potential supply-chain delays and FX volatility.                   |

---

## 🔬 Training Details

* **Base model:** Qwen-2.5B quantized GGUF
* **Environment:** Google Colab, 4-bit quantization for memory efficiency
* **Optimization:** Fine-tuned using Unsloth/PEFT on curated JSONL dataset

---

## 🤝 Acknowledgements

* **Hugging Face & Kaggle** for model hosting and data
* **Open source communities** for technical guidance

---

## 🌐 License

Apache 2.0

---