Artificial Intelligence · 04.08.2026, 07:18 UTC
When Top-K Misses the Decision: Tool-Call Drift in Multi-Teacher On-Policy Distillation
| Schweregrad | info |
|---|---|
| Kategorie | Artificial Intelligence |
| Quelle | arXiv cs.CL ↗ |
| Veröffentlicht | 04.08.2026 UTC |
Sicherheitsmeldung mit Schweregrad noch nicht bewertet. Technische Details im Tab „Originaltext“; empfohlene Schritte in der Checkliste.
arXiv:2607.07050v4 Announce Type: replace Abstract: Top-K teacher logits make on-policy distillation tractable, but probability mass is not decision support. In a two-teacher tool-use setting, vanilla generalized knowledge distillation raises tool-call recall but also over-calls on direct-answer examples. With Qwen3.5-9B, the response teacher's top-32 retains 99.99% of its probability mass yet contains the behavior-switch tool-call token on only 0.4% of 500 unique response prompts; even top-256 covers only 52.2%. Student logits for tokens omitted from the truncated support receive zero direct gradient, so the tool teacher reinforces entry while the response teacher usually cannot oppose it; a wrong entry then amplifies divergence along the generated trajectory. Matched restoration connects this blind spot to behavior: restoring the tool-call token at every supervised response position lowers full-generation over-calling from 14.2+/-2.1% to 3.7+/-0.5% across three seeds, but also lowers call recall by 12.4 points. A non-tool placebo changes over-calling by only 0.95 points, whereas a teacher/student top-32 support-union intervention reaches 7.4+/-0.6% over-calling with 87.0+/-2.0% call recall. The pattern recurs with Llama-3.1-8B's native JSON protocol: the response teacher omits the JSON-entry token through top-256 despite 99.977% retained mass, whereas the tool teacher ranks it first on every audited tool prompt; the matched support-union intervention lowers over-calling from 28.8+/-0.8% to 11.1+/-1.2%. These results identify decision-critical support omission as a …