AI Paper Atlas

Research map / Large language models / Agents, code and math

ControlLLM: Augment Language Models with Tools by Searching on Graphs

Zhaoyang Liu, Zeqiang Lai, Zhangwei Gao, erfei cui, Ziheng Li, Xizhou Zhu, Lewei Lu, Qifeng Chen, Yu Qiao, Jifeng Dai, Wenhai Wang

ECCV 2024 · Large language models › Agents, code and math · approach: Agent instruction-following benchmarks · 60 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
search a pre-built tool graph (Thoughts-on-Graph) instead of LLM-generated chain or tree planningControlLLM · ControlLLM tool benchmarksolution evaluation (All): 0.59 -> 0.93Improved
use beam search instead of greedy search on the tool graph (ablation)ToG · ControlLLM tool benchmarksolution evaluation (All): 0.49 -> 0.88Improved
use adaptive search instead of exhaustive search (ablation)ToG · ControlLLM tool benchmarksolution evaluation (All): 0.97 -> 0.93Mixed
add prior knowledge (likely tools) to subtask description (ablation)ControlLLM task decomposition with ChatGPT-3.5 · ControlLLM tool benchmarksolution evaluation (All): 0.83 -> 0.93Improved
add prior knowledge to subtask description with Llama2-13B (ablation)ControlLLM task decomposition · ControlLLM tool benchmarksolution evaluation (All): 0.68 -> 0.82Improved
use a stronger LLM (GPT-4 instead of Llama2-13B) in task decomposition (ablation)ControlLLM · ControlLLM tool benchmarksolution evaluation (All): 0.82 -> 0.98Improved
finetune LLaMA-7B for task decomposition instead of using ChatGPT zero-shot (Mix vs ChatGPT) (ablation)ControlLLM · ControlLLM tool benchmarksolution evaluation (All): 0.64 -> 0.93Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy