README.md

---
license: llama3.1
base_model: meta-llama/Llama-4-Scout-17B-16E-Instruct
tags:
  - finance
  - lending
  - credit-analysis
  - lora
  - mlx
  - llama-4
language:
  - en
pipeline_tag: text-generation
---

# EVA Financial AI - Llama 4 Scout LoRA Adapters

Fine-tuned LoRA adapters for commercial lending, credit analysis, and financial intelligence.

## Model Details

- **Base Model**: meta-llama/Llama-4-Scout-17B-16E-Instruct (MoE architecture)
- **Training Framework**: MLX (Apple Silicon optimized)
- **Quantization**: 4-bit
- **LoRA Rank**: Default
- **Training Data**: 42,505 examples from financial documents

## Training Data Sources

- Credit risk analysis documents
- Underwriting guidelines
- Loan product specifications
- Financial terminology definitions
- Equipment finance documentation
- Real estate lending guidelines
- SBA loan guidelines
- DSCR loan specifications

## Performance

| Metric | Value |
|--------|-------|
| Initial Val Loss | 2.905 |
| Final Val Loss | ~1.0 |
| Training Iterations | 2000 |
| Trainable Parameters | 91.75M (0.085%) |

## Usage with MLX

```python
from mlx_lm import load, generate

# Load base model with adapters
model, tokenizer = load(
    "mlx-community/meta-llama-Llama-4-Scout-17B-16E-4bit",
    adapter_path="evafiai/eva-llama4-scout-financial-lora"
)

# Generate
prompt = "What is a DSCR loan and what are typical requirements?"
response = generate(model, tokenizer, prompt=prompt, max_tokens=500)
print(response)
```

## Usage with Transformers + PEFT

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

# Load base model
base_model = AutoModelForCausalLM.from_pretrained(
    "meta-llama/Llama-4-Scout-17B-16E-Instruct",
    device_map="auto",
    torch_dtype="auto"
)

# Load LoRA adapters
model = PeftModel.from_pretrained(base_model, "evafiai/eva-llama4-scout-financial-lora")
tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-4-Scout-17B-16E-Instruct")

# Generate
inputs = tokenizer("What is DSCR?", return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=200)
print(tokenizer.decode(outputs[0]))
```

## Capabilities

- **Credit Analysis**: Evaluate creditworthiness, FICO scores, tradelines
- **Loan Matching**: Match borrowers to appropriate lenders and products
- **DSCR Calculations**: Debt Service Coverage Ratio analysis
- **Underwriting**: Guidelines interpretation and documentation requirements
- **Financial Terminology**: Definitions and explanations of industry terms

## Limitations

- Optimized for commercial lending scenarios
- Best used with relevant context/RAG for specific lender products
- May require domain-specific prompting for best results

## License

This model inherits the Llama 3.1 Community License from the base model.

## Citation

```
@misc{eva-financial-ai-2024,
  title={EVA Financial AI - Llama 4 Scout LoRA Adapters},
  author={EVA Financial AI},
  year={2024},
  publisher={Hugging Face},
  url={https://huggingface.co/evafiai/eva-llama4-scout-financial-lora}
}
```