---
license: apache-2.0
base_model: Qwen/Qwen2.5-3B-Instruct
tags:
- finance
- question-answering
- qwen
- fine-tuned
datasets:
- sweatSmile/FinanceQA
language:
- en
---

# Qwen3-4B-Instruct-FinanceQA

This is a fine-tuned version of Qwen3-4B-Instruct trained on financial question-answering data.

## Training Details
- **Base model:** Qwen3-4B-Instruct
- **Dataset:** FinanceQA (3.7k samples)
- **Final training loss:** 0.0652
- **Training epochs:** 2
- **Parameters:** 16.5M trainable (LoRA/QLoRA)

## Usage

```python
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("sweatSmile/Qwen3-4B-Instruct-FinanceQA")
model = AutoModelForCausalLM.from_pretrained("sweatSmile/Qwen3-4B-Instruct-FinanceQA")

# Example usage
prompt = "Context: ARCOTECH Company Name: Arcotech Ltd.\nQuestion: What is the equity share capital?\nAnswer:"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_length=200)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```

## Training Data
The model was trained on financial company data including:
- Equity share capital queries
- Shareholders' funds information
- Financial ratios and metrics
- Company-specific financial data

## Limitations
- Trained specifically on financial QA format
- May not perform well on general conversation
- Should be used for financial information retrieval only

## License
Apache 2.0