{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:U5SKOZYE3IYX3CMXILH7IMIEE3","short_pith_number":"pith:U5SKOZYE","schema_version":"1.0","canonical_sha256":"a764a76704da317d899742cff4310426fa4acf93edb2efc65f12698670b03ea5","source":{"kind":"arxiv","id":"2507.11371","version":1},"attestation_state":"computed","paper":{"title":"Step-wise Policy for Rare-tool Knowledge (SPaRK): Offline RL that Drives Diverse Tool Use in LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.MA"],"primary_cat":"cs.LG","authors_text":"Gabriel Bo, Justin Gu, Koa Chang","submitted_at":"2025-07-15T14:44:29Z","abstract_excerpt":"We present Step-wise Policy for Rare-tool Knowledge (SPaRK), a novel reinforcement learning framework that teaches large language models to explore diverse tool usage patterns beyond conventional high-temperature sampling. Building on recent advances in step-wise reinforcement learning, we introduce a dual-objective reward system that simultaneously optimizes for answer quality and tool diversity, training a Llama-3.1 8B model through offline PPO on synthetically generated trajectories from the MMLU-Pro dataset. Our approach uniquely employs a rarity-first exploitation strategy where a GPT-4o "},"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":"2507.11371","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-07-15T14:44:29Z","cross_cats_sorted":["cs.MA"],"title_canon_sha256":"9339e4fbc71a6c409d4a67c3f8eddecd191cc33e3e9ecb879c6e4db0c74e3738","abstract_canon_sha256":"42daa41b113aa63da792fbd193597ba534a53a007c4184114d64e9afa09c3fdb"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:37:38.018017Z","signature_b64":"PbJKBeLsPHiakaOBtK11hNkvbV158OnobvDRSkRM/xNWh7fRjQDuDyKr5N295y7zGvpyS9azYWGudNsvvCY1DA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a764a76704da317d899742cff4310426fa4acf93edb2efc65f12698670b03ea5","last_reissued_at":"2026-07-05T11:37:38.017485Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:37:38.017485Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Step-wise Policy for Rare-tool Knowledge (SPaRK): Offline RL that Drives Diverse Tool Use in LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.MA"],"primary_cat":"cs.LG","authors_text":"Gabriel Bo, Justin Gu, Koa Chang","submitted_at":"2025-07-15T14:44:29Z","abstract_excerpt":"We present Step-wise Policy for Rare-tool Knowledge (SPaRK), a novel reinforcement learning framework that teaches large language models to explore diverse tool usage patterns beyond conventional high-temperature sampling. Building on recent advances in step-wise reinforcement learning, we introduce a dual-objective reward system that simultaneously optimizes for answer quality and tool diversity, training a Llama-3.1 8B model through offline PPO on synthetically generated trajectories from the MMLU-Pro dataset. Our approach uniquely employs a rarity-first exploitation strategy where a GPT-4o "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.11371","kind":"arxiv","version":1},"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/2507.11371/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":"2507.11371","created_at":"2026-07-05T11:37:38.017548+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.11371v1","created_at":"2026-07-05T11:37:38.017548+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.11371","created_at":"2026-07-05T11:37:38.017548+00:00"},{"alias_kind":"pith_short_12","alias_value":"U5SKOZYE3IYX","created_at":"2026-07-05T11:37:38.017548+00:00"},{"alias_kind":"pith_short_16","alias_value":"U5SKOZYE3IYX3CMX","created_at":"2026-07-05T11:37:38.017548+00:00"},{"alias_kind":"pith_short_8","alias_value":"U5SKOZYE","created_at":"2026-07-05T11:37:38.017548+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/U5SKOZYE3IYX3CMXILH7IMIEE3","json":"https://pith.science/pith/U5SKOZYE3IYX3CMXILH7IMIEE3.json","graph_json":"https://pith.science/api/pith-number/U5SKOZYE3IYX3CMXILH7IMIEE3/graph.json","events_json":"https://pith.science/api/pith-number/U5SKOZYE3IYX3CMXILH7IMIEE3/events.json","paper":"https://pith.science/paper/U5SKOZYE"},"agent_actions":{"view_html":"https://pith.science/pith/U5SKOZYE3IYX3CMXILH7IMIEE3","download_json":"https://pith.science/pith/U5SKOZYE3IYX3CMXILH7IMIEE3.json","view_paper":"https://pith.science/paper/U5SKOZYE","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.11371&json=true","fetch_graph":"https://pith.science/api/pith-number/U5SKOZYE3IYX3CMXILH7IMIEE3/graph.json","fetch_events":"https://pith.science/api/pith-number/U5SKOZYE3IYX3CMXILH7IMIEE3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/U5SKOZYE3IYX3CMXILH7IMIEE3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/U5SKOZYE3IYX3CMXILH7IMIEE3/action/storage_attestation","attest_author":"https://pith.science/pith/U5SKOZYE3IYX3CMXILH7IMIEE3/action/author_attestation","sign_citation":"https://pith.science/pith/U5SKOZYE3IYX3CMXILH7IMIEE3/action/citation_signature","submit_replication":"https://pith.science/pith/U5SKOZYE3IYX3CMXILH7IMIEE3/action/replication_record"}},"created_at":"2026-07-05T11:37:38.017548+00:00","updated_at":"2026-07-05T11:37:38.017548+00:00"}