{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:NJWZPV64IO7ZCTWWIDXSPTV7ZM","short_pith_number":"pith:NJWZPV64","schema_version":"1.0","canonical_sha256":"6a6d97d7dc43bf914ed640ef27cebfcb1814d4ee17dd4408302b30ad9e6db307","source":{"kind":"arxiv","id":"2605.28184","version":1},"attestation_state":"computed","paper":{"title":"Joint Training of Multi-Token Prediction in Reinforcement Learning via Optimal Coefficient Calibration","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Guojun Yin, Jiajun Chai, Lin Chen, Shiming Xiang, Xiaohan Wang, Zili Wang","submitted_at":"2026-05-27T09:07:06Z","abstract_excerpt":"Reinforcement Learning from Verifiable Rewards (RLVR) has emerged as the standard paradigm for improving reasoning capability of large language models, while Multi-Token Prediction (MTP) has been a widely adopted module in pretraining. Combining them is a natural approach, yet current RL practices detach MTP gradients because joint training degrades the performance. We revisit this failure from an optimization perspective. We show that the per-step effect of MTP on the RL objective can be decomposed into two terms: a first-order correlation and a second-order perturbation penalty. This decompo"},"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":"2605.28184","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-27T09:07:06Z","cross_cats_sorted":[],"title_canon_sha256":"6eb05ef761edb4033965c02e201fc0369be4ac1c365f34e90e07eafa9756d20a","abstract_canon_sha256":"a9d22f5560c428abbea72904c78f459e3524918514a2fc58dcc58cab40d27881"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-28T01:05:01.586370Z","signature_b64":"9aUb6kMgfsDKPToqgqHqbzk+QDcPZddIMtuBoHGGQm41dWg9TeH96arUWa9WSuhO+mQVFbbPTn04QbKmCyU2AA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6a6d97d7dc43bf914ed640ef27cebfcb1814d4ee17dd4408302b30ad9e6db307","last_reissued_at":"2026-05-28T01:05:01.585951Z","signature_status":"signed_v1","first_computed_at":"2026-05-28T01:05:01.585951Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Joint Training of Multi-Token Prediction in Reinforcement Learning via Optimal Coefficient Calibration","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Guojun Yin, Jiajun Chai, Lin Chen, Shiming Xiang, Xiaohan Wang, Zili Wang","submitted_at":"2026-05-27T09:07:06Z","abstract_excerpt":"Reinforcement Learning from Verifiable Rewards (RLVR) has emerged as the standard paradigm for improving reasoning capability of large language models, while Multi-Token Prediction (MTP) has been a widely adopted module in pretraining. Combining them is a natural approach, yet current RL practices detach MTP gradients because joint training degrades the performance. We revisit this failure from an optimization perspective. We show that the per-step effect of MTP on the RL objective can be decomposed into two terms: a first-order correlation and a second-order perturbation penalty. This decompo"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.28184","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/2605.28184/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":"2605.28184","created_at":"2026-05-28T01:05:01.586007+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.28184v1","created_at":"2026-05-28T01:05:01.586007+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.28184","created_at":"2026-05-28T01:05:01.586007+00:00"},{"alias_kind":"pith_short_12","alias_value":"NJWZPV64IO7Z","created_at":"2026-05-28T01:05:01.586007+00:00"},{"alias_kind":"pith_short_16","alias_value":"NJWZPV64IO7ZCTWW","created_at":"2026-05-28T01:05:01.586007+00:00"},{"alias_kind":"pith_short_8","alias_value":"NJWZPV64","created_at":"2026-05-28T01:05:01.586007+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/NJWZPV64IO7ZCTWWIDXSPTV7ZM","json":"https://pith.science/pith/NJWZPV64IO7ZCTWWIDXSPTV7ZM.json","graph_json":"https://pith.science/api/pith-number/NJWZPV64IO7ZCTWWIDXSPTV7ZM/graph.json","events_json":"https://pith.science/api/pith-number/NJWZPV64IO7ZCTWWIDXSPTV7ZM/events.json","paper":"https://pith.science/paper/NJWZPV64"},"agent_actions":{"view_html":"https://pith.science/pith/NJWZPV64IO7ZCTWWIDXSPTV7ZM","download_json":"https://pith.science/pith/NJWZPV64IO7ZCTWWIDXSPTV7ZM.json","view_paper":"https://pith.science/paper/NJWZPV64","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.28184&json=true","fetch_graph":"https://pith.science/api/pith-number/NJWZPV64IO7ZCTWWIDXSPTV7ZM/graph.json","fetch_events":"https://pith.science/api/pith-number/NJWZPV64IO7ZCTWWIDXSPTV7ZM/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NJWZPV64IO7ZCTWWIDXSPTV7ZM/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NJWZPV64IO7ZCTWWIDXSPTV7ZM/action/storage_attestation","attest_author":"https://pith.science/pith/NJWZPV64IO7ZCTWWIDXSPTV7ZM/action/author_attestation","sign_citation":"https://pith.science/pith/NJWZPV64IO7ZCTWWIDXSPTV7ZM/action/citation_signature","submit_replication":"https://pith.science/pith/NJWZPV64IO7ZCTWWIDXSPTV7ZM/action/replication_record"}},"created_at":"2026-05-28T01:05:01.586007+00:00","updated_at":"2026-05-28T01:05:01.586007+00:00"}