{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:L2P4S757ZPAEOW3TOZALVQZPZ6","short_pith_number":"pith:L2P4S757","schema_version":"1.0","canonical_sha256":"5e9fc97fbfcbc0475b737640bac32fcfbbb2ef75e0bd05da7906d773599c3ca0","source":{"kind":"arxiv","id":"2604.11557","version":2},"attestation_state":"computed","paper":{"title":"UniToolCall: Unifying Tool-Use Representation, Data, and Evaluation for LLM Agents","license":"http://creativecommons.org/licenses/by/4.0/","headline":"UniToolCall standardizes tool-use data and evaluation so that a fine-tuned 8B model reaches 93 percent precision on complex agent tasks.","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Changyu Zeng, Hao Wu, Wei Xing, Xiaoyu Shen, Xinghao Chen, Yifan Ge, Yijuan Liang, Ziyi Wu","submitted_at":"2026-04-13T14:43:47Z","abstract_excerpt":"Tool-use capability is a fundamental component of LLM agents, enabling them to interact with external systems through structured function calls. However, existing research exhibits inconsistent interaction representations, largely overlooks the structural distribution of tool-use trajectories, and relies on incompatible evaluation benchmarks. We present UniToolCall, a unified framework for tool learning that standardizes the entire pipeline from toolset construction and dataset generation to evaluation. The framework curates a large tool pool of 22k+ tools and constructs a hybrid training corp"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2604.11557","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-04-13T14:43:47Z","cross_cats_sorted":[],"title_canon_sha256":"fc7bc9e123e9af5cce266bca01d6cab06e7991356d33c5761ba10d67419b0f9c","abstract_canon_sha256":"cdb4899a6ad5a029dc43705e8f54e1e0443fd297dc469d093840563cb429cde3"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-26T02:04:10.595824Z","signature_b64":"EHFKQ7GQCZSUdE7DPHWGgCjcH8J6G99lhuD8O21ydfn+nIFlGMlzoDTovt3XIg6J1RPFpLMRgorQTM6rFsOtDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"5e9fc97fbfcbc0475b737640bac32fcfbbb2ef75e0bd05da7906d773599c3ca0","last_reissued_at":"2026-05-26T02:04:10.595138Z","signature_status":"signed_v1","first_computed_at":"2026-05-26T02:04:10.595138Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"UniToolCall: Unifying Tool-Use Representation, Data, and Evaluation for LLM Agents","license":"http://creativecommons.org/licenses/by/4.0/","headline":"UniToolCall standardizes tool-use data and evaluation so that a fine-tuned 8B model reaches 93 percent precision on complex agent tasks.","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Changyu Zeng, Hao Wu, Wei Xing, Xiaoyu Shen, Xinghao Chen, Yifan Ge, Yijuan Liang, Ziyi Wu","submitted_at":"2026-04-13T14:43:47Z","abstract_excerpt":"Tool-use capability is a fundamental component of LLM agents, enabling them to interact with external systems through structured function calls. However, existing research exhibits inconsistent interaction representations, largely overlooks the structural distribution of tool-use trajectories, and relies on incompatible evaluation benchmarks. We present UniToolCall, a unified framework for tool learning that standardizes the entire pipeline from toolset construction and dataset generation to evaluation. The framework curates a large tool pool of 22k+ tools and constructs a hybrid training corp"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"fine-tuning Qwen3-8B on our dataset substantially improves tool-use performance. Under the distractor-heavy Hybrid-20 setting, achieves 93.0% single-turn Strict Precision, outperforming commercial models including GPT, Gemini, and Claude.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That combining public datasets with structurally controlled synthetic trajectories and the Anchor Linkage mechanism produces training data that genuinely improves generalization to real tool-use scenarios without introducing artifacts or biases.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"UniToolCall unifies tool-use data and evaluation for LLM agents, enabling fine-tuned models to reach 93% single-turn precision on a challenging benchmark with distractors.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"UniToolCall standardizes tool-use data and evaluation so that a fine-tuned 8B model reaches 93 percent precision on complex agent tasks.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"c856544bc8314c70c944f2a5400d02f3103cec46110c04fa5a34462f158113d9"},"source":{"id":"2604.11557","kind":"arxiv","version":2},"verdict":{"id":"caa9bdf3-de7b-40ee-b5b8-8f8689cd8c6d","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T15:48:16.551695Z","strongest_claim":"fine-tuning Qwen3-8B on our dataset substantially improves tool-use performance. Under the distractor-heavy Hybrid-20 setting, achieves 93.0% single-turn Strict Precision, outperforming commercial models including GPT, Gemini, and Claude.","one_line_summary":"UniToolCall unifies tool-use data and evaluation for LLM agents, enabling fine-tuned models to reach 93% single-turn precision on a challenging benchmark with distractors.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That combining public datasets with structurally controlled synthetic trajectories and the Anchor Linkage mechanism produces training data that genuinely improves generalization to real tool-use scenarios without introducing artifacts or biases.","pith_extraction_headline":"UniToolCall standardizes tool-use data and evaluation so that a fine-tuned 8B model reaches 93 percent precision on complex agent tasks."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.11557/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":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2604.11557","created_at":"2026-05-26T02:04:10.595232+00:00"},{"alias_kind":"arxiv_version","alias_value":"2604.11557v2","created_at":"2026-05-26T02:04:10.595232+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.11557","created_at":"2026-05-26T02:04:10.595232+00:00"},{"alias_kind":"pith_short_12","alias_value":"L2P4S757ZPAE","created_at":"2026-05-26T02:04:10.595232+00:00"},{"alias_kind":"pith_short_16","alias_value":"L2P4S757ZPAEOW3T","created_at":"2026-05-26T02:04:10.595232+00:00"},{"alias_kind":"pith_short_8","alias_value":"L2P4S757","created_at":"2026-05-26T02:04:10.595232+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/L2P4S757ZPAEOW3TOZALVQZPZ6","json":"https://pith.science/pith/L2P4S757ZPAEOW3TOZALVQZPZ6.json","graph_json":"https://pith.science/api/pith-number/L2P4S757ZPAEOW3TOZALVQZPZ6/graph.json","events_json":"https://pith.science/api/pith-number/L2P4S757ZPAEOW3TOZALVQZPZ6/events.json","paper":"https://pith.science/paper/L2P4S757"},"agent_actions":{"view_html":"https://pith.science/pith/L2P4S757ZPAEOW3TOZALVQZPZ6","download_json":"https://pith.science/pith/L2P4S757ZPAEOW3TOZALVQZPZ6.json","view_paper":"https://pith.science/paper/L2P4S757","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2604.11557&json=true","fetch_graph":"https://pith.science/api/pith-number/L2P4S757ZPAEOW3TOZALVQZPZ6/graph.json","fetch_events":"https://pith.science/api/pith-number/L2P4S757ZPAEOW3TOZALVQZPZ6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/L2P4S757ZPAEOW3TOZALVQZPZ6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/L2P4S757ZPAEOW3TOZALVQZPZ6/action/storage_attestation","attest_author":"https://pith.science/pith/L2P4S757ZPAEOW3TOZALVQZPZ6/action/author_attestation","sign_citation":"https://pith.science/pith/L2P4S757ZPAEOW3TOZALVQZPZ6/action/citation_signature","submit_replication":"https://pith.science/pith/L2P4S757ZPAEOW3TOZALVQZPZ6/action/replication_record"}},"created_at":"2026-05-26T02:04:10.595232+00:00","updated_at":"2026-05-26T02:04:10.595232+00:00"}