Luigi/lfm2.5-350m-cursor-zh
Luigi/lfm2.5-350m-cursor-zh is a 0.35 billion parameter language model developed by Luigi, specifically designed as a CURSOR agent for editing Chinese-Traditional meeting notes. It shares the same architecture and pipeline as its English counterpart but is trained on synthetic zh-TW traces with deliberate oversampling. This model excels at identifying and correcting stale-state bullets in meeting summaries, passing G1 capability screens for decision chains and deadlines.
Loading preview...
LFM2.5-350M CURSOR Agent — Chinese-Traditional Meeting Notes Editor
This model, developed by Luigi, is the Chinese-Traditional variant of the LFM2.5-350M CURSOR agent, designed for editing meeting notes. It utilizes the same underlying architecture and operational pipeline as the English model, Luigi/lfm2.5-350m-cursor-en, but is specialized for Traditional Chinese.
Key Differentiators and Capabilities
- Targeted Language: Specifically trained for Chinese-Traditional (zh-TW) meeting note editing.
- Synthetic Training Data: Trained exclusively on synthetic zh-TW traces, with an emphasis on oversampling to address known asymmetries in revision behavior. Real-world contested Chinese data remains unmeasured.
- Evaluation Performance: Achieved a PASS on the G1 capability screen, demonstrating proficiency in decision chains, deadlines, anchors, and trap scenarios, with 100% valid operations. Evaluation on T1 zh meetings shows that its VERIFY/ANCHOR sweep effectively eliminates stale-state bullets.
- Operational Consistency: Maintains identical operational grammar and harness as the English model, requiring only the
--lang zh-TWsystem prompt for deployment.
Use Cases
This model is ideal for developers requiring an automated agent to process and refine meeting notes in Traditional Chinese, particularly for tasks involving the identification and correction of outdated or incorrect bullet points within summaries.