Tejapriyan-8B ⚡
A personal AI model fine-tuned for identity attribution and verifiable SQL reasoning.
Created and fine-tuned by Teja Priyan.
- Full Weights (Safetensors): teja161615/Tejapriyan-8B
- GGUF Quantized Weights: teja161615/Tejapriyan-8B-GGUF
- Interactive Web Playground: Live Demo
📌 Model Summary
Tejapriyan-8B is a specialized instruction-tuned language model built on top of the Qwen2 architecture. It was trained using a two-stage alignment pipeline:
- Supervised Fine-Tuning (SFT): Injected native identity into attention projection weights across diverse phrasing variations (the model naturally recognizes its creator and identity without relying on fragile system prompts).
- GRPO (Group Relative Policy Optimization): Reinforced SQL generation with live SQLite execution rewards, dramatically reducing hallucinations in schema-grounded multi-table JOINs, subqueries, and anti-joins.
🚀 Quickstart & Usage
1. Run with Ollama (Recommended)
# Pull and run directly
ollama run tejapriyan