{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:EY3PDNJBD6OKFBPHANUNHTKKHN","short_pith_number":"pith:EY3PDNJB","schema_version":"1.0","canonical_sha256":"2636f1b5211f9ca285e70368d3cd4a3b74bdaf56c49274ec4914d04f93acb9c7","source":{"kind":"arxiv","id":"2412.15495","version":2},"attestation_state":"computed","paper":{"title":"TL-Training: A Task-Feature-Based Framework for Training Large Language Models in Tool Use","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Jianping Fan, Junjie Ye, Peng Wang, Qi Zhang, Sixian Li, Tao Gui, Xuanjing Huang, Yilong Wu, Yuming Yang, Zhengyin Du, Zhiheng Xi, Zhongchao Shi","submitted_at":"2024-12-20T02:21:36Z","abstract_excerpt":"Large language models (LLMs) achieve remarkable advancements by leveraging tools to interact with environments, a critical step toward generalized AI. However, the standard supervised fine-tuning (SFT) approach, which relies on large-scale datasets, often overlooks task-specific characteristics in tool use, leading to performance bottlenecks. To address this issue, we analyze three existing LLMs and uncover key insights: training data can inadvertently impede tool-use behavior, token importance is distributed unevenly, and errors in tool calls fall into a small set of categories. Building on t"},"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":"2412.15495","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-12-20T02:21:36Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"04e3d7c3759bee2f4e10144c902ffd642a0f8ee41a0622cc896e9d3d7a8aed9d","abstract_canon_sha256":"971fe54385e44cbe81aeea2caf6df58de28b7ccf6740b875146770530c9d7e6f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:59:09.890454Z","signature_b64":"MTkRc9m8GEWXD/w3gLTDx5gr2DqGzVb5ctGBwAxvrAD+Rmmpak1mtZmJiv+P8qA4wu0By5yo5hRvQxge462LBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2636f1b5211f9ca285e70368d3cd4a3b74bdaf56c49274ec4914d04f93acb9c7","last_reissued_at":"2026-07-05T11:59:09.889968Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:59:09.889968Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"TL-Training: A Task-Feature-Based Framework for Training Large Language Models in Tool Use","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Jianping Fan, Junjie Ye, Peng Wang, Qi Zhang, Sixian Li, Tao Gui, Xuanjing Huang, Yilong Wu, Yuming Yang, Zhengyin Du, Zhiheng Xi, Zhongchao Shi","submitted_at":"2024-12-20T02:21:36Z","abstract_excerpt":"Large language models (LLMs) achieve remarkable advancements by leveraging tools to interact with environments, a critical step toward generalized AI. However, the standard supervised fine-tuning (SFT) approach, which relies on large-scale datasets, often overlooks task-specific characteristics in tool use, leading to performance bottlenecks. To address this issue, we analyze three existing LLMs and uncover key insights: training data can inadvertently impede tool-use behavior, token importance is distributed unevenly, and errors in tool calls fall into a small set of categories. Building on t"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2412.15495","kind":"arxiv","version":2},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2412.15495/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":"2412.15495","created_at":"2026-07-05T11:59:09.890024+00:00"},{"alias_kind":"arxiv_version","alias_value":"2412.15495v2","created_at":"2026-07-05T11:59:09.890024+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2412.15495","created_at":"2026-07-05T11:59:09.890024+00:00"},{"alias_kind":"pith_short_12","alias_value":"EY3PDNJBD6OK","created_at":"2026-07-05T11:59:09.890024+00:00"},{"alias_kind":"pith_short_16","alias_value":"EY3PDNJBD6OKFBPH","created_at":"2026-07-05T11:59:09.890024+00:00"},{"alias_kind":"pith_short_8","alias_value":"EY3PDNJB","created_at":"2026-07-05T11:59:09.890024+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.06387","citing_title":"WebMCP Tool Surface Poisoning: Runtime Manipulation Attacks on LLM Agents","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2505.07591","citing_title":"MulDimIF: A Multi-Dimensional Constraint Framework for Evaluating and Improving Instruction Following in Large Language Models","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2509.18847","citing_title":"Failure Makes the Agent Stronger: Enhancing Accuracy through Structured Reflection for Reliable Tool Interactions","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.11928","citing_title":"When Simulation Lies: A Sim-to-Real Benchmark and Domain-Randomized RL Recipe for Tool-Use Agents","ref_index":49,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/EY3PDNJBD6OKFBPHANUNHTKKHN","json":"https://pith.science/pith/EY3PDNJBD6OKFBPHANUNHTKKHN.json","graph_json":"https://pith.science/api/pith-number/EY3PDNJBD6OKFBPHANUNHTKKHN/graph.json","events_json":"https://pith.science/api/pith-number/EY3PDNJBD6OKFBPHANUNHTKKHN/events.json","paper":"https://pith.science/paper/EY3PDNJB"},"agent_actions":{"view_html":"https://pith.science/pith/EY3PDNJBD6OKFBPHANUNHTKKHN","download_json":"https://pith.science/pith/EY3PDNJBD6OKFBPHANUNHTKKHN.json","view_paper":"https://pith.science/paper/EY3PDNJB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2412.15495&json=true","fetch_graph":"https://pith.science/api/pith-number/EY3PDNJBD6OKFBPHANUNHTKKHN/graph.json","fetch_events":"https://pith.science/api/pith-number/EY3PDNJBD6OKFBPHANUNHTKKHN/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EY3PDNJBD6OKFBPHANUNHTKKHN/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EY3PDNJBD6OKFBPHANUNHTKKHN/action/storage_attestation","attest_author":"https://pith.science/pith/EY3PDNJBD6OKFBPHANUNHTKKHN/action/author_attestation","sign_citation":"https://pith.science/pith/EY3PDNJBD6OKFBPHANUNHTKKHN/action/citation_signature","submit_replication":"https://pith.science/pith/EY3PDNJBD6OKFBPHANUNHTKKHN/action/replication_record"}},"created_at":"2026-07-05T11:59:09.890024+00:00","updated_at":"2026-07-05T11:59:09.890024+00:00"}