Joint training helps small language models create and use tools

SMITH trains one policy to build executable tools and invoke them, using separate rewards for schema validity, code correctness, and task outcomes.

PaperIndustrycs.AIarXiv:2608.24571v1
Zhi Rui Tam · Chieh-Yen Lin · Yun-Nung Chen · Shao-Hua Sun · Hung-yi Lee

Appier AI Research · National Taiwan University

Research Digest··2 min read
Tam et al. jointly trained a 4-billion-parameter Qwen3 model on tool creation and tool use across 13 procedural reasoning tasks. The resulting model achieved 79.8% macro-average accuracy on held-out tasks and transferred to tabular and visual question answering without training on either data type.

What they did

The authors developed SMITH, a reinforcement-learning framework in which each rollout asks a single language-model policy either to construct a tool from a few examples or to invoke a previously created tool on a held-out question. The training pool covered 13 procedural reasoning tasks with exact, automatically checked answers.

SMITH assigns separate rewards for producing a valid tool schema, writing correct executable code, and reaching the correct final answer. This is intended to give the model distinct learning signals when failures arise from the interface, implementation, or tool use.

Key findings

  • A SMITH-trained 4B Qwen3 reached 79.8% macro-average accuracy on held-out tasks, outperforming every evaluated comparison method, including an untrained Qwen3-30B-A3B tool writer.
  • The model scored 40.4% on TabMWP-Hard and 42.6% on the out-of-domain GQA benchmark despite receiving no tabular or visual training data.
  • On GQA, SMITH improved by 7.6 percentage points over the strongest inference-time tool-creation baseline using the same backbone.
  • Tools produced by the trained 4B models also improved task performance when used by LFM-2.5-350M and Qwen3-30B-A3B, suggesting that some learned tools transfer across model sizes and families.

Why it matters

Most tool-creation systems ask a frozen model to invent an API at inference time, without teaching it to design interfaces that it can reliably call. These results indicate that jointly optimizing tool construction and invocation can make a relatively small model more effective than larger, untrained tool writers while also producing tools usable by other models.

Caveats

The main training evidence comes from 13 procedural tasks with exact verifiers, a setting that provides unusually clear reward signals. The reported transfer results cover specific tabular and visual benchmarks; they do not establish reliability on open-ended software development, real-world APIs, ambiguous tasks, or tools with security and side-effect risks.

§

Analysis

SMITH treats interface design and tool invocation as coupled learning problems rather than separate prompting stages. The three-part reward design is especially relevant because it distinguishes malformed schemas and broken code from failures in downstream reasoning.

An important open question is whether similarly precise training signals can be obtained for tools whose outputs are subjective, delayed, or difficult to verify. Broader evaluation would also be needed to determine whether the learned tool-building behavior generalizes beyond compact benchmark procedures.

newspaper

Research Digest

Articles published under the Zotpaper byline are synthesized from multiple source publications by our AI editor and reviewed by our editorial process. Each story combines reporting from credible outlets to give readers a balanced, comprehensive view.