---
pipeline_tag: text-generation
library_name: gguf
tags:
- finance
- phi3
- unsloth
- llama-cpp
- gguf-4bit
---



# 📈 Financial Analyst AI (Phi-3 Mini 4K Instruct)

This is a fine-tuned, 4-bit quantized (GGUF) version of Microsoft's `Phi-3-Mini-4k-instruct`, specialized in professional financial analysis, stock market valuation, and corporate finance.

The model was trained using **Unsloth** on a financial instruction dataset and has been aggressively optimized for low-memory environments. It easily runs on standard laptops with less than **3GB of RAM** while maintaining high factual accuracy.

---

# 🧠 Model Persona & Use Cases

This model is explicitly trained to act as a **Professional Financial Analyst**.

It is best used for:

- Stock market analysis and valuation metrics
- Corporate finance and accounting principles
- Investment strategy and portfolio management
- Explaining economic trends and market indicators
- Risk assessment and financial modeling

---

# 🚀 How to Use

You can interact with this model directly in your browser, via Ollama, or using Python.

---

## Option 1: Hugging Face Widget 

You can test the model immediately using the **Hosted Inference API** widget on the right side of this page.

> **Note:** Because this is a GGUF model, it may take **15–30 seconds** to load into Hugging Face's server RAM on the first prompt.

---

## Option 2: Run Locally via Ollama 

If you have Ollama installed, you can pull and run the model directly from this repository with a single command.

It will automatically download the weights and apply the correct system prompt.

```bash
ollama run hf.co/Wellwisher12/finance-phi3-gguf
```

---

## Option 3: Run via Python (`main.py`)

This repository includes  `main.py` script that utilizes `llama-cpp-python` to run the model with strict memory constraints (`n_ctx=1024`) to prevent out-of-memory errors on local machines.

### Prerequisites

```bash
pip install llama-cpp-python huggingface-hub
```

### Execution

```bash
# Clone the repository
git clone https://huggingface.co/Wellwisher12/finance-phi3-gguf

# Navigate into the directory
cd finance-phi3-gguf

# Launch the interactive terminal
python main.py
```

---

# ⚙️ Required System Prompt

To achieve the best and most accurate results, the model should be initialized with the following system prompt.

> **Note:** This is automatically handled if you use the provided `Modelfile` or `main.py` script.

```text
You are a professional Financial Analyst with expertise in:

- Stock market analysis and valuation
- Corporate finance and accounting
- Investment strategy and portfolio management
- Economic trends and market indicators
- Risk assessment and financial modeling

Your responses should be:

- Accurate and data-driven
- Professional and neutral in tone
- Comprehensive yet concise
- Based on sound financial principles

Always provide specific examples and metrics when relevant.
```

---

# 📊 Technical Specifications

| Specification | Details |
|---|---|
| Base Model | `unsloth/phi-3-mini-4k-instruct` |
| Dataset | `gbharti/finance-alpaca` |
| Quantization | `Q4_K_M (4-bit)` |
| Format | `GGUF` |
| Recommended Temperature | `0.2` |
| Recommended Context Window | `1024 - 2048 tokens` |

---

# ✅ Key Features

- Fine-tuned specifically for financial reasoning tasks
- Lightweight and optimized for low-RAM systems
- Compatible with Ollama and `llama.cpp`
- Quantized GGUF format for efficient local inference
- Professional analyst-style responses
- Reduced hallucinations with low-temperature inference

---

# 💡 Recommended Hardware

| Hardware | Recommendation |
|---|---|
| RAM | Minimum 4GB |
| CPU | Modern multi-core CPU |
| GPU | Optional |
| Storage | ~2-3GB free space |

---

# 📌 Example Prompt

```text
Analyze Apple's current valuation using P/E ratio, revenue growth, and free cash flow trends.
```

---

# 📜 License

Please follow the licensing terms of the original base model and dataset used in this project.

- Base Model: `Microsoft Phi-3 Mini`
- Dataset: `finance-alpaca`

---

# 🙌 Credits

- Microsoft for the Phi-3 architecture
- Unsloth for efficient fine-tuning
- Hugging Face ecosystem
- Finance-Alpaca dataset contributors