Llama-4-Scout-Finance-Ascended

This model is a specialized financial intelligence engine optimized for zero-shot quantitative reasoning, context extraction, and regulatory compliance tailored specifically to the Indian economic ecosystem.

Built for the Adaption AutoScientist Challenge × HackIndia 2026 by Team Ascended.

📊 Performance Breakthroughs

• Quality Improvement: +51.7% relative quality improvement over the foundational baseline model.
• Domain Win Rate: Achieved a 70% win rate on complex corporate and financial QA datasets.
• Evaluation Grade: Upgraded from a baseline 'C' to a production-ready 'A'.

🧬 Pipeline & Recipe Details

• Base Model: meta-llama/Llama-4-Scout-17B-16E-Instruct
• Training Framework: Supervised Fine-Tuning (SFT) via LoRA (r=64, alpha=256)
• Dataset Used: mishface123/adaption-econ-finance-qa-pairs
• Dataset Mutation: 5,000 raw financial reasoning pairs evolved using Adaption's Adaptive Data engine with prompt rephrasing, metadata injection, and 10% target localization for Indian market parameters.