Shivacon-Energy: The 0.5B Energy LLM That Refuses to Hallucinate
TL;DR: Shivacon-Energy is a LoRA fine-tune of Qwen2.5-0.5B-Instruct for energy-domain Q&A (solar, wind, storage, grids) that passed a brutal 4-gate sellability harness: 9/10 domain tasks, 5/5 unknown-honesty, 5/5 refusal, 5/5 JSON formatting, 0/10 hallucinations. It was the only survivor of 9 domain adapters trained on a free Colab T4 — and the repo publishes the 8 failures too. Public at…
Shivacon-Energy is a fine-tuned language model that refuses to hallucinate when answering energy-related questions. Developed from the Qwen2.5-0.5B-Instruct model, it passed a rigorous 4-gate sellability test, scoring highly in domain tasks, refusal to generate unverified information, JSON formatting, and refusing to emit false data.
The model's primary goal is to provide accurate information without inventing facts or making mistakes in tool output formats. The repository includes all necessary components for using and auditing the model, such as weights, adapters, tokenizer configuration, and evaluation reports. The training process utilized a small dataset with LoRA (Low-Rank Adaptation) on a free Google Colab T4 GPU, incurring no compute costs.
The 4-gate sellability harness evaluated the model against nine different domains, with only Shivacon-Energy successfully completing all gates. The model's limitations are clearly documented, emphasizing that it does not possess additional knowledge beyond the base model and that it is only effective within the energy domain. Despite minor issues like occasional garbled responses, Shivacon-Energy is considered sellable, providing a valuable tool for energy-related queries while maintaining integrity in its outputs.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
