Moving WeOutside246 from GPT-5 to local models on a base M4 Mac Mini

Based on your detailed analysis of various models for an event extraction task, several key insights stand out:

Model Performance Highlights

1. Gemma 4 26B A4B:

   • Best Accuracy: 96%
   • Perfect Recall: No false negatives (missed no real events)
   • False Positives: 8
   • Conclusion: This model is ideal for minimizing the risk of missing actual events, which is crucial for ensuring completeness in your product.

2. Qwen 3.5 9B:

   • Best Precision: 94.7%
   • False Positives: Only 2
   • Conclusion: This model excels at minimizing false positives but may miss some true events, making it suitable if avoiding incorrect classifications is a priority.

3. Qwen 3.5 35B A3B:

   • Best Judge Score: Average score of 70.3
   • Conclusion: Produces more polished and coherent structured outputs but may have slightly worse event detection accuracy compared to Gemma 4 26B.

Fine-Tuning Insights

• The fine-tuned

Read the full article at DEV Community

---

Want to create content about this topic? Use Nemati AI tools to generate articles, social posts, and more.