README.md

---
license: apache-2.0
base_model: mlx-community/Llama-3.2-3B-Instruct-4bit
tags:
  - finance
  - lender-matching
  - mlx
  - lora
  - dpo
  - sba-loans
language:
  - en
pipeline_tag: text-generation
---

# EVA Lender Matching Model

Fine-tuned Llama 3.2 3B model for SBA lender matching and financial advisory.

## Model Description

This model was trained using:
- **SFT (Supervised Fine-Tuning)**: 500 iterations on lender matching data
- **DPO (Direct Preference Optimization)**: 500 iterations for preference alignment

### Training Results
| Stage | Initial Loss | Final Loss | Improvement |
|-------|-------------|------------|-------------|
| Val Loss | 2.902 | 0.331 | 88.6% reduction |
| Train Loss | 2.495 | 0.300 | 88.0% reduction |

## Usage

```python
from mlx_lm import load, generate

model, tokenizer = load(
    "mlx-community/Llama-3.2-3B-Instruct-4bit",
    adapter_path="evafiai/eva-lender-matching"
)

prompt = "I need a $500,000 SBA loan for my manufacturing business. What lenders do you recommend?"
response = generate(model, tokenizer, prompt=prompt, max_tokens=300)
print(response)
```

## Capabilities

- SBA 7(a) and 504 loan recommendations
- Lender matching based on business profile
- NAICS code identification
- Industry-specific financing guidance

## Training Data

Trained on:
- 7,393 preference pairs for DPO
- Comprehensive lender database with verified contacts
- NAICS code matching examples

## License

Apache 2.0