{"paper":{"title":"ToolGrad: Efficient Tool-use Dataset Generation with Textual \"Gradients\"","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"ToolGrad generates superior tool-use training data by building valid chains first with textual gradients before creating queries.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Haoyu Zhang, Jingtao Zhou, Kohei Uehara, Lin Gu, Ruofei Du, Tatsuya Harada, Zheng Xu, Zhongyi Zhou","submitted_at":"2025-08-06T05:04:00Z","abstract_excerpt":"Prior work synthesizes tool-use LLM datasets by first generating a user query, followed by complex tool-use annotations like depth-first search (DFS). This leads to inevitable annotation failures and low efficiency in data generation. We introduce ToolGrad, an agentic framework that inverts this paradigm. ToolGrad first constructs valid tool-use chains through an iterative process guided by textual \"gradients\", and then synthesizes corresponding user queries. This \"answer-first\" approach led to ToolGrad-500, a dataset generated with more complex tool use, lower cost, and almost 100% pass rate."},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Experiments show that ToolGrad models outperform those trained on expensive baseline datasets and proprietary LLMs.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the iterative process guided by textual gradients consistently produces valid and complex tool-use chains that generalize well to real user queries.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"ToolGrad inverts the standard tool-use dataset synthesis process by constructing valid tool chains first with textual gradients, producing a high-quality 500-example dataset with near-perfect validity and superior model performance.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"ToolGrad generates superior tool-use training data by building valid chains first with textual gradients before creating queries.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"fd9909475f388fdb1765e9721d6ff5b5f3f52037511fef0a19cef40651da75b7"},"source":{"id":"2508.04086","kind":"arxiv","version":3},"verdict":{"id":"04a25205-0fe6-4b58-a662-fc914a67685c","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T01:07:59.573950Z","strongest_claim":"Experiments show that ToolGrad models outperform those trained on expensive baseline datasets and proprietary LLMs.","one_line_summary":"ToolGrad inverts the standard tool-use dataset synthesis process by constructing valid tool chains first with textual gradients, producing a high-quality 500-example dataset with near-perfect validity and superior model performance.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the iterative process guided by textual gradients consistently produces valid and complex tool-use chains that generalize well to real user queries.","pith_extraction_headline":"ToolGrad generates superior tool-use training data by building valid chains first with textual gradients before creating queries."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2508.04086/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":1,"snapshot_sha256":"daae8f819e1046ec10cddd5ed7bc9e6dc291025fa7e8edb1146e9b3246d4f5c0"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}