Research map / Large language models / Agents, code and math
CATP-LLM: Empowering Large Language Models for Cost-Aware Tool Planning
Duo Wu, Jinghe Wang, Yuan Meng, Yanning Zhang, Le Sun, Zhi Wang
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add context augmentation for cost-aware features (ablation) | CATP-LLM · OpenCATP sequential | QoP: 0.246 -> 0.261 | Improved |
| add intermediate reward feedback at each action (ablation) | CATP-LLM · OpenCATP sequential | QoP: 0.209 -> 0.261 | Improved |
| add adaptive masking of invalid tool/dependency tokens (ablation) | CATP-LLM · OpenCATP sequential | valid plans: 70.2% -> 100% | Improved |
| use CATP-LLM (Llama2-7B) instead of GPT-based methods | tool planning · OpenCATP sequential | QoP: +1.5%-40.6% QoP on average | Improved |
| use CATP-LLM instead of RLTF (final-only feedback) | Llama2-7B fine-tuning · OpenCATP sequential | QoP: 0.181 -> 0.261 | Improved |
| use CATP-LLM instead of GPT and fine-tuning baselines | non-sequential planning · OpenCATP non-sequential | QoP: 92.0%-375.4% higher QoP on average | Improved |
| use tool planning language with CATP-LLM for non-sequential plans | Llama2-7B · OpenCATP non-sequential | valid plans: GPT-3.5 invalid 41.7%, 75%, 66.7% vs 100% valid | Improved |
| scale backbone from smaller models to Llama2-7B (ablation) | CATP-LLM · OpenCATP non-sequential | QoP | Improved |
| RLTF vs CATP-LLM on task score | sequential planning · OpenCATP sequential | Task score: 0.715 (RLTF) vs 0.681 | Mixed |
Closest papers by meaning
- ToolTree: Efficient LLM Tool Planning via Dual-Feedback Monte Carlo Tree Search and Bidirectional Pruning (ICLR 2026 · similarity 0.88)
- Tool-Planner: Task Planning with Clusters across Multiple Tools (ICLR 2025 · similarity 0.88)
- CRAFT: Customizing LLMs by Creating and Retrieving from Specialized Toolsets (ICLR 2024 · similarity 0.88)
- Large Language Models as Tool Makers (ICLR 2024 · similarity 0.87)
- Joint Optimization of Tool Creation and Use for Large Language Model Agents (NeurIPS 2026 · similarity 0.87)
- Thought of Search: Planning with Language Models Through The Lens of Efficiency (NeurIPS 2024 · similarity 0.87)
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.
