VittoriaLanzo/Ohmatic-Qwen3-8B

Ohmatic-Qwen3-8B: Self-Correcting Circuit Schematic Generation

Ohmatic-Qwen3-8B is an 8-billion parameter model, fully merged from Qwen3-8B, engineered to generate electronic circuit schematics from natural language. What sets it apart is its innovative closed verification loop, which enables it to not only generate schematics but also to self-correct them based on Electrical Rule Checker (ERC) feedback.

Key Capabilities

• Natural Language to Schematic: Translates plain language circuit descriptions into structured schematics (components, values, nets).
• Integrated ERC Verification: Schematics are automatically checked against a deterministic Electrical Rule Checker for issues like shorts, floating nets, and polarity errors.
• Self-Correction Mechanism: Upon ERC failure, the model receives feedback and generates a repaired schematic, improving designs rather than simply re-rolling them. This correction turn is explicitly taught during training.
• Fully Merged Weights: The model weights are fully merged, meaning no PEFT/adapter is required; it loads like any causal language model.

Training and Performance

The model was trained to both produce circuits and repair its own designs using only ERC-verified examples. At selection time, it achieved a 53.1% first-pass validity on held-out ERC tests, with the correction loop recovering additional failures. The training data, recipe, and ERC engine are proprietary.

Good for

• Automated generation of electronic circuit schematics.
• Applications requiring robust, error-checked circuit designs.
• Integrating circuit design capabilities into larger systems where self-correction is beneficial.