{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:6R467E6TVIVI5JO5FIWQIO2VKZ","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"f204d9fca2c9e83bd116974f0d6aabf6e4729352a52a7fa4639d51c7340caec4","cross_cats_sorted":["cs.AI","cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-05-19T10:14:08Z","title_canon_sha256":"329ba22ef4faf0002e9306b72aae413a0174ef713c8411654eb9db536cc3396c"},"schema_version":"1.0","source":{"id":"2505.12929","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2505.12929","created_at":"2026-07-05T11:05:11Z"},{"alias_kind":"arxiv_version","alias_value":"2505.12929v1","created_at":"2026-07-05T11:05:11Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.12929","created_at":"2026-07-05T11:05:11Z"},{"alias_kind":"pith_short_12","alias_value":"6R467E6TVIVI","created_at":"2026-07-05T11:05:11Z"},{"alias_kind":"pith_short_16","alias_value":"6R467E6TVIVI5JO5","created_at":"2026-07-05T11:05:11Z"},{"alias_kind":"pith_short_8","alias_value":"6R467E6T","created_at":"2026-07-05T11:05:11Z"}],"graph_snapshots":[{"event_id":"sha256:c3857af420b0b2a057279d9fb274dcdbb37fe9f24b53f93627e502d0e27043c6","target":"graph","created_at":"2026-07-05T11:05:11Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2505.12929/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Reinforcement learning (RL) has become a cornerstone for enhancing the reasoning capabilities of large language models (LLMs), with recent innovations such as Group Relative Policy Optimization (GRPO) demonstrating exceptional effectiveness. In this study, we identify a critical yet underexplored issue in RL training: low-probability tokens disproportionately influence model updates due to their large gradient magnitudes. This dominance hinders the effective learning of high-probability tokens, whose gradients are essential for LLMs' performance but are substantially suppressed. To mitigate th","authors_text":"Dongqi Han, Dongsheng Li, Xufang Luo, Yunjian Xu, Zhihe Yang, Zhiyuan He, Zilong Wang","cross_cats":["cs.AI","cs.LG"],"headline":"","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-05-19T10:14:08Z","title":"Do Not Let Low-Probability Tokens Over-Dominate in RL for LLMs"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.12929","kind":"arxiv","version":1},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:360d4bed5f934a57361fdd0dc0f981b50a346dbf97ebd98849a49820bbc9abc2","target":"record","created_at":"2026-07-05T11:05:11Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"f204d9fca2c9e83bd116974f0d6aabf6e4729352a52a7fa4639d51c7340caec4","cross_cats_sorted":["cs.AI","cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-05-19T10:14:08Z","title_canon_sha256":"329ba22ef4faf0002e9306b72aae413a0174ef713c8411654eb9db536cc3396c"},"schema_version":"1.0","source":{"id":"2505.12929","kind":"arxiv","version":1}},"canonical_sha256":"f479ef93d3aa2a8ea5dd2a2d043b555643f976b92504a26248f4ec34e550b015","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"f479ef93d3aa2a8ea5dd2a2d043b555643f976b92504a26248f4ec34e550b015","first_computed_at":"2026-07-05T11:05:11.993626Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T11:05:11.993626Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"nyYbXTFoD0mvd1kP5j20mMQxwltld2el78Nsy3H7XiWqZRaX7Jkf46w/rSj3T+S7eVW1HLJcojufFPogRl97DQ==","signature_status":"signed_v1","signed_at":"2026-07-05T11:05:11.994113Z","signed_message":"canonical_sha256_bytes"},"source_id":"2505.12929","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:360d4bed5f934a57361fdd0dc0f981b50a346dbf97ebd98849a49820bbc9abc2","sha256:c3857af420b0b2a057279d9fb274dcdbb37fe9f24b53f93627e502d0e27043c6"],"state_sha256":"8aed8cff1a012fa45f4c3c0272b8155092f3e470195e422b0a201f6445f56a5f"}