{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:JBBVU5HSDXAMALWEIU76ZKLN43","short_pith_number":"pith:JBBVU5HS","schema_version":"1.0","canonical_sha256":"48435a74f21dc0c02ec4453feca96de6e080f361226c02f8467c24ab6eca1a4e","source":{"kind":"arxiv","id":"2507.19017","version":1},"attestation_state":"computed","paper":{"title":"MindSpeed RL: Distributed Dataflow for Scalable and Efficient RL Training on Ascend NPU Cluster","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Beirong Zhou, Benzhe Ning, Bo Wang, Chang Liu, Chenyi Pan, Fei Mei, Guang Yang, Jiangben Wang, Jianxiang Zhang, Laingjun Feng, Xinjie Guo, Xinyang Liu, Zeng Shu, Zhenyu Han","submitted_at":"2025-07-25T07:11:49Z","abstract_excerpt":"Reinforcement learning (RL) is a paradigm increasingly used to align large language models. Popular RL algorithms utilize multiple workers and can be modeled as a graph, where each node is the status of a worker and each edge represents dataflow between nodes. Owing to the heavy cross-node dependencies, the RL training system usually suffers from poor cluster scalability and low memory utilization. In this article, we introduce MindSpeed RL, an effective and efficient system for large-scale RL training. Unlike existing centralized methods, MindSpeed RL organizes the essential data dependencies"},"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":"2507.19017","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-07-25T07:11:49Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"379d4fdf35ab7110f668339844ca2f6c70dd3a174531907a21e9bb542bcda883","abstract_canon_sha256":"13d2e4328a716e499a9bf76c57754a84606318f97f1aa74094b53500b742d02a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:43:18.014972Z","signature_b64":"rNDI0kEpBp/3S10wciWpEZVV+M36dMGJ4J+YivIbOawaIsNfApBXBp3+7oLTkp7qjwuupWlk0a9dYciPcRFVDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"48435a74f21dc0c02ec4453feca96de6e080f361226c02f8467c24ab6eca1a4e","last_reissued_at":"2026-07-05T11:43:18.014488Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:43:18.014488Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"MindSpeed RL: Distributed Dataflow for Scalable and Efficient RL Training on Ascend NPU Cluster","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Beirong Zhou, Benzhe Ning, Bo Wang, Chang Liu, Chenyi Pan, Fei Mei, Guang Yang, Jiangben Wang, Jianxiang Zhang, Laingjun Feng, Xinjie Guo, Xinyang Liu, Zeng Shu, Zhenyu Han","submitted_at":"2025-07-25T07:11:49Z","abstract_excerpt":"Reinforcement learning (RL) is a paradigm increasingly used to align large language models. Popular RL algorithms utilize multiple workers and can be modeled as a graph, where each node is the status of a worker and each edge represents dataflow between nodes. Owing to the heavy cross-node dependencies, the RL training system usually suffers from poor cluster scalability and low memory utilization. In this article, we introduce MindSpeed RL, an effective and efficient system for large-scale RL training. Unlike existing centralized methods, MindSpeed RL organizes the essential data dependencies"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.19017","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/2507.19017/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":"2507.19017","created_at":"2026-07-05T11:43:18.014546+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.19017v1","created_at":"2026-07-05T11:43:18.014546+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.19017","created_at":"2026-07-05T11:43:18.014546+00:00"},{"alias_kind":"pith_short_12","alias_value":"JBBVU5HSDXAM","created_at":"2026-07-05T11:43:18.014546+00:00"},{"alias_kind":"pith_short_16","alias_value":"JBBVU5HSDXAMALWE","created_at":"2026-07-05T11:43:18.014546+00:00"},{"alias_kind":"pith_short_8","alias_value":"JBBVU5HS","created_at":"2026-07-05T11:43:18.014546+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.06111","citing_title":"Schedule-and-Calibrate: Utility-Guided Multi-Task Reinforcement Learning for Code LLMs","ref_index":35,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/JBBVU5HSDXAMALWEIU76ZKLN43","json":"https://pith.science/pith/JBBVU5HSDXAMALWEIU76ZKLN43.json","graph_json":"https://pith.science/api/pith-number/JBBVU5HSDXAMALWEIU76ZKLN43/graph.json","events_json":"https://pith.science/api/pith-number/JBBVU5HSDXAMALWEIU76ZKLN43/events.json","paper":"https://pith.science/paper/JBBVU5HS"},"agent_actions":{"view_html":"https://pith.science/pith/JBBVU5HSDXAMALWEIU76ZKLN43","download_json":"https://pith.science/pith/JBBVU5HSDXAMALWEIU76ZKLN43.json","view_paper":"https://pith.science/paper/JBBVU5HS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.19017&json=true","fetch_graph":"https://pith.science/api/pith-number/JBBVU5HSDXAMALWEIU76ZKLN43/graph.json","fetch_events":"https://pith.science/api/pith-number/JBBVU5HSDXAMALWEIU76ZKLN43/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/JBBVU5HSDXAMALWEIU76ZKLN43/action/timestamp_anchor","attest_storage":"https://pith.science/pith/JBBVU5HSDXAMALWEIU76ZKLN43/action/storage_attestation","attest_author":"https://pith.science/pith/JBBVU5HSDXAMALWEIU76ZKLN43/action/author_attestation","sign_citation":"https://pith.science/pith/JBBVU5HSDXAMALWEIU76ZKLN43/action/citation_signature","submit_replication":"https://pith.science/pith/JBBVU5HSDXAMALWEIU76ZKLN43/action/replication_record"}},"created_at":"2026-07-05T11:43:18.014546+00:00","updated_at":"2026-07-05T11:43:18.014546+00:00"}