ProgAgent: Learning Tool Orchestration through Programmatic Training

Hugging Face Daily Papers Papers
llm-agents tool-orchestration programmatic-calling training-framework long-horizon-tasks reinforcement-learning

Summary

ProgAgent 提出一个完整的程序化工具调用（PTC）训练框架，通过可验证的长时程任务合成、轻量级 PTC 运行时和两阶段训练配方，显著优于同规模基线，在 LOCA-bench 64k token 状态下达 30.7% 通过率而基线为零。

Large language models (LLMs) are increasingly expected to orchestrate diverse tools over long-horizon tasks. Existing agents typically use Direct Tool Calling (DTC), where the model invokes tools sequentially, and tool responses accumulate in the model context, which can lead to context rot and orchestration breakdown. In contrast, Programmatic Tool Calling (PTC) invokes tools through executable code and stores intermediate results in runtime variables rather than appending them to the context. Prior work has explored PTC mainly as an inference-time optimization, but its benefits for agentic training remain unclear, and suitable verifiable training tasks are scarce. In this work, we introduce ProgAgent, a full PTC training stack, comprising a pipeline for synthesizing verifiable long-horizon tasks, a lightweight PTC runtime, and a two-stage training recipe. Using only 215 tasks instantiated at four environment scales, ProgAgent yields higher pass rates than the evaluated baselines of comparable size across four benchmarks. On LOCA-bench, ProgAgent-14B achieves a 30.7% pass rate at a 64k-token environment state, where these baselines score zero. Cross-harness experiments show that PTC-trained models outperform DTC-trained models under both evaluation interfaces, suggesting that the benefits of PTC training extend beyond the interface.
Original Article
View Cached Full Text

Cached at: 10/15/26, 07:02 AM

Paper page - ProgAgent: Learning Tool Orchestration through Programmatic Training

Source: https://huggingface.co/papers/2610.14774

Abstract

Largelanguagemodels(LLMs)areincreasinglyexpectedtoorchestratediversetoolsoverlong-horizontasks.ExistingagentstypicallyuseDirectToolCalling(DTC),wherethemodelinvokestoolssequentially,andtoolresponsesaccumulateinthemodelcontext,whichcanleadtocontextrotandorchestrationbreakdown.Incontrast,ProgrammaticToolCalling(PTC)invokestoolsthroughexecutablecodeandstoresintermediateresultsinruntimevariablesratherthanappendingthemtothecontext.PriorworkhasexploredPTCmainlyasaninference-timeoptimization,butitsbenefitsforagentictrainingremainunclear,andsuitableverifiabletrainingtasksarescarce.Inthiswork,weintroduceProgAgent,afullPTCtrainingstack,comprisingapipelineforsynthesizingverifiablelong-horizontasks,alightweightPTCruntime,andatwo-stagetrainingrecipe.Usingonly215tasksinstantiatedatfourenvironmentscales,ProgAgentyieldshigherpassratesthantheevaluatedbaselinesofcomparablesizeacrossfourbenchmarks.OnLOCA-bench,ProgAgent-14Bachievesa30.7%passrateata64k-tokenenvironmentstate,wherethesebaselinesscorezero.Cross-harnessexperimentsshowthatPTC-trainedmodelsoutperformDTC-trainedmodelsunderbothevaluationinterfaces,suggestingthatthebenefitsofPTCtrainingextendbeyondtheinterface.

View arXiv pageView PDFProject pageGitHubAdd to collection

Get this paper in your agent:

hf papers read 2610\.14774

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2610.14774 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2610.14774 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2610.14774 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Tool-Making and Self-Evolving LLM Agents in Low-Latency Systems

arXiv cs.CL

This paper presents a method for compiling repeated standard operating procedure steps into validated, versioned tools before deployment, replacing inference-time code generation. In a fulfillment center alarm-triage system, this approach reduces p50 latency by 42% and end-to-end error rate by up to 53%.

Speculative Programmatic Tool Calling (12 minute read)

TLDR AI

The article proposes speculative programmatic tool calling (sPTC), a technique to optimize tool calls in AI harnesses by overlapping execution with token generation, reducing latency inspired by speculative execution in CPUs and LLMs.

@mylifcc: Major Experiment: Using LLM as an 'Optimization Agent' for Automatic Loop Scheduling! Just read this paper accepted at PACT 2025: 'Agentic Auto-Scheduling: An Experimental Study of LLM-Guided Loop Opt…'

X AI KOLs Timeline

Introduces a paper accepted at PACT 2025, proposing the ComPilot framework, which uses off-the-shelf LLMs as optimization agents to automatically optimize complex loop nests without fine-tuning, achieving a geometric mean speedup of 3.54x, surpassing the SOTA Pluto.