WEFT: Scaling Tool-Use Post-Training for General-Purpose Agents
WEFT scales full agent systems for tool-use post-training and beats size-matched baselines on major agent benchmarks.
WEFT jointly scales agent environments, tasks, harnesses, and evaluators for tool-use post-training instead of scaling environments alone. Execution-driven self-evolution attributes failures from traces and revises the responsible component, while prefix-preserving sampling, atomic-turn credit assignment, and MegaMCP stabilize large concurrent training. WEFT-8B and WEFT-14B beat matched environment-scaling baselines on BFCL V4, τ²-Bench, and Claw-Eval. WEFT-14B improves over Agent-World-14B by 6.41, 2.23, and 12.27 points, and WEFT-35B-A3B extends gains to Toolathlon-Verified and AutomationBench.
- Scales environments, tasks, harnesses, and evaluators together.
- Self-evolution revises failing components using execution traces.
- MegaMCP isolates recoverable state across concurrent tool rollouts.
- WEFT-14B beats Agent-World-14B by up to 12.27 points.
- Products
- MegaMCP
From the original article
1 min read at huggingface.co
Recent efforts to scale tool-use post-training have largely centered on the synthesis of executable environments, which constitute only one component of a broader agentic interaction system comprising the environment, task, agent harness, and evaluator.
By Bo Mao, Hang He, Linting Wang · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.