Tool Retrieval and Function Selection on Taskbench-MM
27Function Selection AccuracyDynamic LR (DTDR-L)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Dynamic LR (DTDR-L)LLM=Qwen 3 0.6B, ICL Method=hard masking2025.12 | 27 | 0.69 | 0.55 | |
| Dynamic DR (DTDR-C)LLM=Qwen 3 0.6B, ICL Method=hard masking2025.12 | 24.4 | 0.52 | 0.43 | |
| LRLLM=Qwen 3 0.6B, ICL Method=hard masking2025.12 | 21 | 0.49 | 0.28 | |
| BM-25LLM=Qwen 3 0.6B, ICL Method=hard masking2025.12 | 13.7 | 0.26 | 0.19 | |
| DRLLM=Qwen 3 0.6B, ICL Method=hard masking2025.12 | 13.3 | 0.38 | 0.27 | |
| QTS (Paramanayakam et al. 2025)LLM=Qwen 3 0.6B, ICL Method=hard masking2025.12 | 7.6 | 0.29 | 0.18 | |
| QTS (Vanilla)LLM=Qwen 3 0.6B, ICL Method=hard masking2025.12 | 5.3 | 0.14 | 0.04 | |
| QTS (Gao et al. 2025)LLM=Qwen 3 0.6B, ICL Method=hard masking2025.12 | 5.3 | 0.14 | 0.2 | |
| Random GuessLLM=Qwen 3 0.6B, ICL Method=hard masking2025.12 | 2.4 | 0.11 | 0.03 |