Truthseeker87/solarhive-26b-a4b-merged

SolarHive 26B A4B Merged: Community Solar Energy Intelligence

SolarHive 26B A4B Merged is a production-ready, LoRA fine-tuned Gemma 4 26B A4B (MoE) model developed by Truthseeker87. It is specifically designed for community solar energy intelligence, offering domain expertise in solar production, battery management, grid optimization, and community coordination. This model integrates native function calling for four energy-specific tools and supports multimodal visual question answering (VQA) for tasks like sky analysis and panel inspection.

Key Capabilities

• Domain Expertise: Specialized in community solar energy, including solar production, battery management, and grid optimization.
• Multimodal VQA: Analyzes sky photographs for cloud cover, panel images for damage/shading, and aerial images for panel inventory.
• Native Function Calling: Equipped with 5 tools (get_weather, get_solar_production, get_battery_state, get_grid_status, get_nrel_pvwatts_baseline) for agentic workflows, grounding responses with real API data.
• Selective Tool Reasoning: Intelligently decides when and which tools to call based on the query, validated by When2Call probes.
• Optimized Architecture: Utilizes Gemma 4 26B A4B's MoE architecture (25.2B total, 3.8B active parameters) with a ~550M vision encoder and 256K context window, offering a strong capability-to-cost ratio for cloud inference.

Good For

• Developers building AI energy advisors for community microgrids and storage systems.
• Applications requiring intelligent analysis of solar data, weather impacts, and grid conditions.
• Integrating AI with external energy APIs for grounded, data-driven responses.
• Scenarios demanding multimodal input processing (text and images) in the energy sector.