{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:DWO2LGIPCGJBGRAXWESWP6ZDQ3","short_pith_number":"pith:DWO2LGIP","schema_version":"1.0","canonical_sha256":"1d9da5990f1192134417b12567fb2386f0968d92947f89c1f944f486f62d9f18","source":{"kind":"arxiv","id":"2512.06547","version":4},"attestation_state":"computed","paper":{"title":"A-3PO: Accelerating Asynchronous LLM Training with Staleness-aware Proximal Policy Approximation","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI","cs.DC"],"primary_cat":"cs.LG","authors_text":"Shiliang Wu, Xiaocan Li, Zheng Shen","submitted_at":"2025-12-06T19:37:39Z","abstract_excerpt":"Decoupled PPO has been a successful reinforcement learning (RL) algorithm to deal with the high data staleness under the asynchronous RL setting. Decoupled loss used in decoupled PPO improves coupled-loss style of algorithms' (e.g., standard PPO, GRPO) learning stability by introducing a proximal policy to decouple the off-policy correction (importance weight) from the policy update constraint (trust region). However, the proximal policy requires an extra forward pass through the model at each training step, creating a computational overhead for large language models training. We observe that "},"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":"2512.06547","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2025-12-06T19:37:39Z","cross_cats_sorted":["cs.AI","cs.DC"],"title_canon_sha256":"4e6d57915d838058ce480edd740a3cc06a7e6a251982874dc2fa29b91be52a16","abstract_canon_sha256":"2708f2941fe89118537e43b1a26c5fa288c84b67c6256c7303541aade6c0aac9"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-13T01:29:39.212770Z","signature_b64":"zP7jeXnNFFc5IPT897Y9EL+/4MvFON90csn8AKHWMeGniMVLmlHTk+HY15Op8pezLdCT0pZi7Hkn8oGEg2TdBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1d9da5990f1192134417b12567fb2386f0968d92947f89c1f944f486f62d9f18","last_reissued_at":"2026-08-13T01:29:39.210165Z","signature_status":"signed_v1","first_computed_at":"2026-08-13T01:29:39.210165Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"A-3PO: Accelerating Asynchronous LLM Training with Staleness-aware Proximal Policy Approximation","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":["cs.AI","cs.DC"],"primary_cat":"cs.LG","authors_text":"Shiliang Wu, Xiaocan Li, Zheng Shen","submitted_at":"2025-12-06T19:37:39Z","abstract_excerpt":"Decoupled PPO has been a successful reinforcement learning (RL) algorithm to deal with the high data staleness under the asynchronous RL setting. Decoupled loss used in decoupled PPO improves coupled-loss style of algorithms' (e.g., standard PPO, GRPO) learning stability by introducing a proximal policy to decouple the off-policy correction (importance weight) from the policy update constraint (trust region). However, the proximal policy requires an extra forward pass through the model at each training step, creating a computational overhead for large language models training. We observe that "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2512.06547","kind":"arxiv","version":4},"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/2512.06547/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":"2512.06547","created_at":"2026-08-13T01:29:39.211321+00:00"},{"alias_kind":"arxiv_version","alias_value":"2512.06547v4","created_at":"2026-08-13T01:29:39.211321+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2512.06547","created_at":"2026-08-13T01:29:39.211321+00:00"},{"alias_kind":"pith_short_12","alias_value":"DWO2LGIPCGJB","created_at":"2026-08-13T01:29:39.211321+00:00"},{"alias_kind":"pith_short_16","alias_value":"DWO2LGIPCGJBGRAX","created_at":"2026-08-13T01:29:39.211321+00:00"},{"alias_kind":"pith_short_8","alias_value":"DWO2LGIP","created_at":"2026-08-13T01:29:39.211321+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":3,"sample":[{"citing_arxiv_id":"2606.24143","citing_title":"AsyncOPD: How Stale Can On-Policy Distillation Be?","ref_index":10,"is_internal_anchor":true},{"citing_arxiv_id":"2606.03382","citing_title":"Local Guidance, Global Impact: Gaussian-Reshaped Trust Region Unlocks Behavior Transitions","ref_index":27,"is_internal_anchor":true},{"citing_arxiv_id":"2605.12070","citing_title":"Missing Old Logits in Asynchronous Agentic RL: Semantic Mismatch and Repair Methods for Off-Policy Correction","ref_index":12,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/DWO2LGIPCGJBGRAXWESWP6ZDQ3","json":"https://pith.science/pith/DWO2LGIPCGJBGRAXWESWP6ZDQ3.json","graph_json":"https://pith.science/api/pith-number/DWO2LGIPCGJBGRAXWESWP6ZDQ3/graph.json","events_json":"https://pith.science/api/pith-number/DWO2LGIPCGJBGRAXWESWP6ZDQ3/events.json","paper":"https://pith.science/paper/DWO2LGIP"},"agent_actions":{"view_html":"https://pith.science/pith/DWO2LGIPCGJBGRAXWESWP6ZDQ3","download_json":"https://pith.science/pith/DWO2LGIPCGJBGRAXWESWP6ZDQ3.json","view_paper":"https://pith.science/paper/DWO2LGIP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2512.06547&json=true","fetch_graph":"https://pith.science/api/pith-number/DWO2LGIPCGJBGRAXWESWP6ZDQ3/graph.json","fetch_events":"https://pith.science/api/pith-number/DWO2LGIPCGJBGRAXWESWP6ZDQ3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DWO2LGIPCGJBGRAXWESWP6ZDQ3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DWO2LGIPCGJBGRAXWESWP6ZDQ3/action/storage_attestation","attest_author":"https://pith.science/pith/DWO2LGIPCGJBGRAXWESWP6ZDQ3/action/author_attestation","sign_citation":"https://pith.science/pith/DWO2LGIPCGJBGRAXWESWP6ZDQ3/action/citation_signature","submit_replication":"https://pith.science/pith/DWO2LGIPCGJBGRAXWESWP6ZDQ3/action/replication_record"}},"created_at":"2026-08-13T01:29:39.211321+00:00","updated_at":"2026-08-13T01:29:39.211321+00:00"}