Soft Clamp cuts tool-call overuse in multi-teacher distillation, from 13.7% to 9.0%
antgroup · hf · 2026-07-21
What the paper says
Multi-teacher on-policy distillation for agentic language models can improve tool-call recall, but the paper finds it also shifts behavior toward over-calling: the student starts invoking tools in cases that should be answered directly.
- The authors argue that aggregate losses can miss this boundary drift.
- They analyze the problem as behavior leverage imbalance: token-level signals near mode-entry points such as <toolcall> and function names can dominate the global generation mode.
- To address this, they introduce Soft Clamp, a per-token divergence calibration method that compresses extreme Jensen-Shannon divergence while keeping gradients nonzero.
Results
- On APIGen-MT, Soft Clamp reduces over-calling from 13.7% to 9.0% relative to vanilla GKD while matching decision accuracy.
- On a BFCL multi-turn diagnostic, it also reduces tool-call loops and repeated calls among GKD variants.
The main takeaway is that multi-teacher OPD should monitor not just how much teacher signal there is, but where it acts during generation.
More from coding & agent
- Cognition's SWE-2 uses a KKT duality argument in RL to shift the effort Pareto curve — YouJiacheng · 2026-09-11
- First-ever Three.js Conference lands in Paris, with a panel on AI-shortened design workflows — OdinLovis · 2026-09-11
- Data engineering, not agent frameworks, is the real bottleneck for enterprise AI agents — dhruv2038 · 2026-09-11
- RTK Terminal Compression Cuts Tokens but Leaves Your AI Coding Bill Unchanged — Bartaseth · 2026-09-11
- GPT-6 Astra beats Factorio with enemies in 44 in-game hours at ~$4,500 API cost — liminal_bardo · 2026-09-11
- Investment Analyst Asks How to Build a Claude-Based Diligence Agent Stack — Careless_Tie2286 · 2026-09-11