This model is a specialized fine-tune of Llama 3.1 8B, designed for senior decision-makers in the energy sector. It translates complex regulatory language into actionable strategic alerts.
🎯 Project Purpose
In the energy market, a single regulation (like a CERC order) has opposite impacts on different stakeholders. This model uses Persona-Aware Instruction Tuning to provide divergent advice for:
Power Generators (Sellers): Focus on revenue certainty, evacuation risks, and project bankability.
Utilities/Discoms (Buyers): Focus on cost pass-through, grid reliability, and tariff impacts.
📚 Training Data & Sources
The model was fine-tuned on a high-density dataset consisting of:
CERC Orders (2024-25): Technical rulings on ISTS licenses, GNA, and grid code compliance (e.g., Petition No. 513/TL/2024).
IEA World Energy Outlook 2025: Strategic macro-trends including AI-driven load growth and global decarbonization pathways.
🛠 Technical Specifications
Architecture: Llama-3.1 8B
Fine-Tuning Method: QLoRA (4-bit quantization)
Optimization: Unsloth (2x faster training)
Quantization: GGUF (q4_k_m) for local deployment.
🚀 Example Prompt
Instruction: Analyze this clause as a Utility/Discom: "The transmission license is subject to annual fee payments."
Response: * Strategic Impact: License fees are a recurring O&M expense.
Financial Risk: Medium (Tariff pass-through dependency).
Action Item: Ensure these fees are included in the upcoming ARR filing to the SERC.
