{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:SGTFOMT5U2N567NN2QIFNSYAUR","short_pith_number":"pith:SGTFOMT5","schema_version":"1.0","canonical_sha256":"91a657327da69bdf7dadd41056cb00a44df0832c7bc88a360b85393174efae65","source":{"kind":"arxiv","id":"2602.15620","version":5},"attestation_state":"computed","paper":{"title":"STAPO: Stabilizing Reinforcement Learning for LLMs by Silencing Rare Spurious Tokens","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"Silencing gradients from a tiny fraction of spurious tokens stabilizes RL fine-tuning of LLMs and raises math reasoning performance.","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Bo Zhang, Guojian Zhan, Jiang Wu, Jingliang Duan, Kehua Sheng, Keqiang Li, Letian Tao, Shengbo Eben Li, Shiqi Liu, Yang Guan, Yinuo Wang, Zeyu He, Zhilong Zheng","submitted_at":"2026-02-17T14:46:48Z","abstract_excerpt":"Reinforcement Learning (RL) has significantly improved large language model reasoning, but existing RL fine-tuning methods rely heavily on heuristic techniques such as entropy regularization and reweighting to maintain stability. In practice, they often suffer from late-stage performance collapse, leading to degraded reasoning quality and unstable training. We identify a key factor behind this instability: a small fraction of tokens, termed spurious tokens (around 0.01%), which contribute little to the reasoning outcome but receive disproportionately amplified gradient updates due to inheritin"},"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":"2602.15620","kind":"arxiv","version":5},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.CL","submitted_at":"2026-02-17T14:46:48Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"b046897e425ba8f1ea06345a01ddc9783375117ae852d676dc9c4e1e7e9e70f1","abstract_canon_sha256":"34b2f2138427fe72e67dea67a30b5e4808a9e4c70f93b18271adfbbecc62bd0b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-26T02:04:06.526733Z","signature_b64":"TYrD7aZy9AvYdRNyunqUtG/2NHDDfiyR8s3xrS0Pmw57iBoLH5ula2ARHpnbDen5D+rGSA5Ya8UHyhMThQ7dBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"91a657327da69bdf7dadd41056cb00a44df0832c7bc88a360b85393174efae65","last_reissued_at":"2026-05-26T02:04:06.525769Z","signature_status":"signed_v1","first_computed_at":"2026-05-26T02:04:06.525769Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"STAPO: Stabilizing Reinforcement Learning for LLMs by Silencing Rare Spurious Tokens","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"Silencing gradients from a tiny fraction of spurious tokens stabilizes RL fine-tuning of LLMs and raises math reasoning performance.","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Bo Zhang, Guojian Zhan, Jiang Wu, Jingliang Duan, Kehua Sheng, Keqiang Li, Letian Tao, Shengbo Eben Li, Shiqi Liu, Yang Guan, Yinuo Wang, Zeyu He, Zhilong Zheng","submitted_at":"2026-02-17T14:46:48Z","abstract_excerpt":"Reinforcement Learning (RL) has significantly improved large language model reasoning, but existing RL fine-tuning methods rely heavily on heuristic techniques such as entropy regularization and reweighting to maintain stability. In practice, they often suffer from late-stage performance collapse, leading to degraded reasoning quality and unstable training. We identify a key factor behind this instability: a small fraction of tokens, termed spurious tokens (around 0.01%), which contribute little to the reasoning outcome but receive disproportionately amplified gradient updates due to inheritin"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Across six mathematical reasoning benchmarks using Qwen 1.7B, 8B, and 14B base models, STAPO consistently demonstrates superior entropy stability and achieves an average performance improvement of 11.49% (ρ_T=1.0, top-p=1.0) and 3.73% (ρ_T=0.7, top-p=0.9) over GRPO, 20-Entropy, and JustRL.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the identified spurious tokens (0.01% fraction) are the primary driver of instability and that silencing their gradients does not discard useful reasoning signal or introduce new biases in the policy update.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"STAPO stabilizes RL for LLMs by suppressing gradient updates from rare spurious tokens, yielding 11.49% average gains on math benchmarks over GRPO and similar baselines.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Silencing gradients from a tiny fraction of spurious tokens stabilizes RL fine-tuning of LLMs and raises math reasoning performance.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"5997830b62c025d9e4f9ea7945a5a9e00fdb91584d71bdb6f762a1928bef4e77"},"source":{"id":"2602.15620","kind":"arxiv","version":5},"verdict":{"id":"28f91ef8-1e4e-447e-a0f7-066f663d1db5","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T21:38:10.962118Z","strongest_claim":"Across six mathematical reasoning benchmarks using Qwen 1.7B, 8B, and 14B base models, STAPO consistently demonstrates superior entropy stability and achieves an average performance improvement of 11.49% (ρ_T=1.0, top-p=1.0) and 3.73% (ρ_T=0.7, top-p=0.9) over GRPO, 20-Entropy, and JustRL.","one_line_summary":"STAPO stabilizes RL for LLMs by suppressing gradient updates from rare spurious tokens, yielding 11.49% average gains on math benchmarks over GRPO and similar baselines.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the identified spurious tokens (0.01% fraction) are the primary driver of instability and that silencing their gradients does not discard useful reasoning signal or introduce new biases in the policy update.","pith_extraction_headline":"Silencing gradients from a tiny fraction of spurious tokens stabilizes RL fine-tuning of LLMs and raises math reasoning performance."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2602.15620/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":"2602.15620","created_at":"2026-05-26T02:04:06.525915+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.15620v5","created_at":"2026-05-26T02:04:06.525915+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.15620","created_at":"2026-05-26T02:04:06.525915+00:00"},{"alias_kind":"pith_short_12","alias_value":"SGTFOMT5U2N5","created_at":"2026-05-26T02:04:06.525915+00:00"},{"alias_kind":"pith_short_16","alias_value":"SGTFOMT5U2N567NN","created_at":"2026-05-26T02:04:06.525915+00:00"},{"alias_kind":"pith_short_8","alias_value":"SGTFOMT5","created_at":"2026-05-26T02:04:06.525915+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":2,"sample":[{"citing_arxiv_id":"2605.22263","citing_title":"Tailoring Teaching to Aptitude: Direction-Adaptive Self-Distillation for LLM Reasoning","ref_index":34,"is_internal_anchor":true},{"citing_arxiv_id":"2605.19425","citing_title":"When to Stop Reusing: Dynamic Gradient Gating for Sample-Efficient RLVR","ref_index":17,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SGTFOMT5U2N567NN2QIFNSYAUR","json":"https://pith.science/pith/SGTFOMT5U2N567NN2QIFNSYAUR.json","graph_json":"https://pith.science/api/pith-number/SGTFOMT5U2N567NN2QIFNSYAUR/graph.json","events_json":"https://pith.science/api/pith-number/SGTFOMT5U2N567NN2QIFNSYAUR/events.json","paper":"https://pith.science/paper/SGTFOMT5"},"agent_actions":{"view_html":"https://pith.science/pith/SGTFOMT5U2N567NN2QIFNSYAUR","download_json":"https://pith.science/pith/SGTFOMT5U2N567NN2QIFNSYAUR.json","view_paper":"https://pith.science/paper/SGTFOMT5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.15620&json=true","fetch_graph":"https://pith.science/api/pith-number/SGTFOMT5U2N567NN2QIFNSYAUR/graph.json","fetch_events":"https://pith.science/api/pith-number/SGTFOMT5U2N567NN2QIFNSYAUR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SGTFOMT5U2N567NN2QIFNSYAUR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SGTFOMT5U2N567NN2QIFNSYAUR/action/storage_attestation","attest_author":"https://pith.science/pith/SGTFOMT5U2N567NN2QIFNSYAUR/action/author_attestation","sign_citation":"https://pith.science/pith/SGTFOMT5U2N567NN2QIFNSYAUR/action/citation_signature","submit_replication":"https://pith.science/pith/SGTFOMT5U2N567NN2QIFNSYAUR/action/replication_record"}},"created_at":"2026-05-26T02:04:06.525915+00:00","updated_at":"2026-05-26T02:04:06.525915+00:00"}