Diamond Swarm - VA Claims AI Agents

Fine-tuned Qwen2.5-7B models specialized for VA disability claims assistance.

Models

Base Model

• Qwen2.5-7B-Instruct - Fine-tuned with QLoRA (4-bit)

Training Data

• 38 CFR Part 4 (VA Rating Schedule)
• BVA case decisions
• VA claims procedures and terminology
• Nexus letter templates
• Personal statement examples

Usage

HuggingFace Transformers

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("Vet-Rate-org/Diamond-Swarm-Auditor-7B")
tokenizer = AutoTokenizer.from_pretrained("Vet-Rate-org/Diamond-Swarm-Auditor-7B")

llama.cpp

# Download GGUF
wget https://huggingface.co/Vet-Rate-org/Diamond-Swarm-Auditor-7B-GGUF/resolve/main/vetrate-auditor-7b-v2-Q4_K_M.gguf

# Run server
./llama-server -m vetrate-auditor-7b-v2-Q4_K_M.gguf -c 4096 -ngl 99 --port 8080

Ollama

# Create Modelfile
FROM ./vetrate-auditor-7b-v2-Q4_K_M.gguf
PARAMETER temperature 0.7
SYSTEM You are a VA claims expert specializing in legal analysis.

# Create model
ollama create diamond-auditor -f Modelfile
ollama run diamond-auditor

GGUF Quantizations

• Q4_K_M (~4.4GB) - Best balance of speed and quality
• Q5_K_M (~5.1GB) - Higher quality, slightly slower

License

MIT - Free for personal and commercial use

Disclaimer

These models provide educational information only. They are NOT legal or medical advice. Always consult with accredited VSOs, attorneys, or medical professionals for official guidance.

Credits

• Built by Vet-Rate.org
• Powered by Qwen2.5