Tool-Need Prediction on Nemotron held-out core (test)
75.3Tool-Need AccuracyGPT-OSS 20B
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GPT-OSS 20BInstruction-tuned (IT)=true, Dominant SAE layers=23, 19, Number of SAE features=200, Probe regularization=lasso2026.05 | 75.3 | 75 | |
| Gemma 3 27BInstruction-tuned (IT)=true, Dominant SAE layers=40, 53, 31, Number of SAE features=2000, Probe regularization=elastic net2026.05 | 71.4 | 71 |