A 3B model with few-shot prompting reaches 79.7% of GPT-5 tool-use performance while a hypernetwork adaptation adds zero measurable benefit across four benchmarks.
Computing Research Repository , volume =
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Meta-Tool: Efficient Few-Shot Tool Adaptation for Small Language Models
A 3B model with few-shot prompting reaches 79.7% of GPT-5 tool-use performance while a hypernetwork adaptation adds zero measurable benefit across four benchmarks.