Back to AI News
arxiv

JET: Just-Enough Exploration Tuning for Efficient Tool-Calling Agents in Repository-Level QA

Researchers identify over-exploration in repository-level QA agents, where LLM agents keep invoking tools after they have gathered enough information to answer. They propose Just-Enough Exploration Tuning (JET), which truncates training trajectories at the sufficiency point, adds a rationale explaining why the agent can answer, and applies higher loss weight to that rationale. Experiments with two open-weight 8B models on two repository-level QA benchmarks show 11–33% fewer tool calls on self-generated trajectories and 18–28% fewer than SFT on expert trajectories, while largely preserving task scores and lowering inference cost.

AI Brief

It offers a concrete fine-tuning technique for reducing unnecessary tool calls and inference cost in repository-level QA agents without discarding most of the performance gain from SFT.

arXiv:2611.22539v1 Announce Type: cross Abstract: In repository-level question answering (QA), an agent explores a code repository with tools and should answer a question once enough information is gathered. Nevertheless, we find that LLM agents often continue to invoke tools even after the gathered information is already sufficient to answer. We call this phenomenon over-exploration. Moreover, even though existing methods improve task performance through supervised fine-tuning (SFT) on expert trajectories, SFT also propagates the over-exploration in the trajectories to the agent. To mitigate over-exploration, we propose Just-Enough Exploration Tuning (JET), a fine-tuning method that trains the agent to answer once the gathered information is sufficient. JET truncates each training trajectory at the step where the gathered information becomes sufficient, inserts a rationale that explains why the agent can answer at that step, and fine-tunes the agent with a higher loss weight on the rationale. Experiments on two repository-level QA benchmarks with two open-weight 8B models demonstrate that JET reduces tool calls by 11 to 33% on self-generated trajectories while largely preserving the score. On expert trajectories, JET makes 18 to 28% fewer tool calls than SFT while retaining most of the score gain of SFT. In both cases, JET also lowers the inference cost.
Be the first to rate this article.

Comments & Questions (0)

Captcha
Please be respectful — let's keep the conversation friendly.

No comments yet

Be the first to comment!