Skill-selection on TOOLBENCH-60 soft-0.3 preference regime
264.8RegretBandit-as-Override
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Bandit-as-OverrideMethod Category=LLM with Statistical Prior, Backbone=Qwen3-30B-Instruct2026.06 | 264.8 | 46.2 | 0.539 | |
| Pure-BanditMethod Category=Statistical, Backbone=Qwen3-30B-Instruct2026.06 | 282 | 39.5 | 0.539 | |
| Freq-as-OverrideMethod Category=LLM with Statistical Prior, Backbone=Qwen3-30B-Instruct2026.06 | 295.3 | 41.6 | 0.288 | |
| Freq-GreedyMethod Category=Statistical, Backbone=Qwen3-30B-Instruct2026.06 | 328.5 | 33.3 | 0.399 | |
| Profile-MemoryMethod Category=LLM with Memory, Backbone=Qwen3-30B-Instruct2026.06 | 344.2 | 32.9 | 0.271 | |
| Bandit-as-ContextMethod Category=LLM with Memory, Backbone=Qwen3-30B-Instruct2026.06 | 369.6 | 26.4 | 0.373 | |
| InContext-MemoryMethod Category=LLM with Memory, Backbone=Qwen3-30B-Instruct2026.06 | 372.7 | 25.4 | 0.282 | |
| ZeroShot-LLMMethod Category=No Learning, Backbone=Qwen3-30B-Instruct2026.06 | 377.5 | 24.3 | 0 | |
| RandomMethod Category=No Learning, Backbone=Qwen3-30B-Instruct2026.06 | 491.2 | 1.7 | 0 |