[KongchangAI]
ConceptJust-Enough Exploration Tuning

JET

A fine-tuning method for repository-level question-answering agents. It truncates trajectories at the point where information is sufficient and inserts rationales with high loss weights, training the agent to stop calling tools once enough information has been gathered and thereby mitigating over-exploration.

Timeline (last 90 days)

Nov 18

The paper "JET: Just-Enough Exploration Tuning for Efficient Tool-Calling Agents in Repository-Level Question Answering" proposes a fine-tuning method named JET, aimed at training agents to stop exploring and answer the question when information is sufficient.

Unverified50%
Nov 18

The JET method truncates training trajectories: it stops at the step where information becomes sufficient and inserts a rationale explaining why to stop.

Unverified50%
Nov 18

During fine-tuning, JET assigns a higher loss weight to the inserted rationale to strengthen the agent's learning of the stopping condition.

Unverified50%
Nov 18

Through truncation and rationale augmentation, JET essentially introduces an emphasis on key decisions within the SFT framework, thereby alleviating SFT's shortcomings.

Unverified50%
Nov 18

JET requires no environment interaction and achieves efficiency optimization solely through supervisory signals.

Unverified50%
Nov 18

The paper conducted experiments on two repository-level question answering benchmarks using two open-source 8-billion-parameter models.

Unverified50%
Nov 18

On self-generated trajectories, JET reduces the number of tool calls by 11% to 33% while largely maintaining task scores.

Unverified50%
Nov 18

In comparison with SFT, JET achieves 18% to 28% fewer tool calls than SFT on expert trajectories while retaining most of SFT's score improvement.

Unverified50%
Nov 18

JET's effectiveness in reducing tool calls is consistent across different models and benchmarks, demonstrating generalization ability.

Unverified50%
Nov 18

JET's method may be generalizable to other scenarios requiring multi-step tool calls, such as web navigation and API calling.

Unverified50%

All Facts (10)

Unverified

The paper "JET: Just-Enough Exploration Tuning for Efficient Tool-Calling Agents in Repository-Level Question Answering" proposes a fine-tuning method named JET, aimed at training agents to stop exploring and answer the question when information is sufficient.

50%
Unverified

The JET method truncates training trajectories: it stops at the step where information becomes sufficient and inserts a rationale explaining why to stop.

50%
Unverified

During fine-tuning, JET assigns a higher loss weight to the inserted rationale to strengthen the agent's learning of the stopping condition.

50%
Unverified

Through truncation and rationale augmentation, JET essentially introduces an emphasis on key decisions within the SFT framework, thereby alleviating SFT's shortcomings.

50%
Unverified

JET requires no environment interaction and achieves efficiency optimization solely through supervisory signals.

50%
Unverified

The paper conducted experiments on two repository-level question answering benchmarks using two open-source 8-billion-parameter models.

50%
Unverified

On self-generated trajectories, JET reduces the number of tool calls by 11% to 33% while largely maintaining task scores.

50%
Unverified

In comparison with SFT, JET achieves 18% to 28% fewer tool calls than SFT on expert trajectories while retaining most of SFT's score improvement.

50%
Unverified

JET's effectiveness in reducing tool calls is consistent across different models and benchmarks, demonstrating generalization ability.

50%
Unverified

JET's method may be generalizable to other scenarios requiring multi-step tool calls, such as web navigation and API calling.

50%

Source Articles