{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:F5FWAOGQMOMNKWULB3ENVPAJIZ","short_pith_number":"pith:F5FWAOGQ","schema_version":"1.0","canonical_sha256":"2f4b6038d06398d55a8b0ec8dabc094669021f455f2ba8bff945c17da3fe0bad","source":{"kind":"arxiv","id":"2507.01663","version":1},"attestation_state":"computed","paper":{"title":"AsyncFlow: An Asynchronous Streaming RL Framework for Efficient LLM Post-Training","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Ansheng You, Chenyi Pan, Chunsheng Li, Chunshi Deng, Guang Yang, Haibo Wang, Huazhong Ji, Jianping Wu, Jing Wang, Kui Luo, Menglong Chen, Sicheng Zhang, Wenjie Liu, Wenqi Shi, Xin Huang, Yixiang Zhang, Yu Huang, Zeshun Lan, Zhenyu Han","submitted_at":"2025-07-02T12:45:34Z","abstract_excerpt":"Reinforcement learning (RL) has become a pivotal technology in the post-training phase of large language models (LLMs). Traditional task-colocated RL frameworks suffer from significant scalability bottlenecks, while task-separated RL frameworks face challenges in complex dataflows and the corresponding resource idling and workload imbalance. Moreover, most existing frameworks are tightly coupled with LLM training or inference engines, making it difficult to support custom-designed engines. To address these challenges, we propose AsyncFlow, an asynchronous streaming RL framework for efficient p"},"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":"2507.01663","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2025-07-02T12:45:34Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"401aeffd81c8bbfc300064943d66524ecdbb609926f6343356fead09d92bb835","abstract_canon_sha256":"a9e6031ed0e2c92aa950d79c25e245fb9ed0fa14f0edf6bee9ff845f719f2b9a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:30:52.806936Z","signature_b64":"FELwJDS/Ws6ZruUYjHd2Rt8l4K3TAoD7Dtdyr1W5gaOq50v46GYnS5BoFmSjJE7WDIImtI+mSzNmLg1JEn7mBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2f4b6038d06398d55a8b0ec8dabc094669021f455f2ba8bff945c17da3fe0bad","last_reissued_at":"2026-07-05T11:30:52.806427Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:30:52.806427Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"AsyncFlow: An Asynchronous Streaming RL Framework for Efficient LLM Post-Training","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Ansheng You, Chenyi Pan, Chunsheng Li, Chunshi Deng, Guang Yang, Haibo Wang, Huazhong Ji, Jianping Wu, Jing Wang, Kui Luo, Menglong Chen, Sicheng Zhang, Wenjie Liu, Wenqi Shi, Xin Huang, Yixiang Zhang, Yu Huang, Zeshun Lan, Zhenyu Han","submitted_at":"2025-07-02T12:45:34Z","abstract_excerpt":"Reinforcement learning (RL) has become a pivotal technology in the post-training phase of large language models (LLMs). Traditional task-colocated RL frameworks suffer from significant scalability bottlenecks, while task-separated RL frameworks face challenges in complex dataflows and the corresponding resource idling and workload imbalance. Moreover, most existing frameworks are tightly coupled with LLM training or inference engines, making it difficult to support custom-designed engines. To address these challenges, we propose AsyncFlow, an asynchronous streaming RL framework for efficient p"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.01663","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/2507.01663/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":"2507.01663","created_at":"2026-07-05T11:30:52.806489+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.01663v1","created_at":"2026-07-05T11:30:52.806489+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.01663","created_at":"2026-07-05T11:30:52.806489+00:00"},{"alias_kind":"pith_short_12","alias_value":"F5FWAOGQMOMN","created_at":"2026-07-05T11:30:52.806489+00:00"},{"alias_kind":"pith_short_16","alias_value":"F5FWAOGQMOMNKWUL","created_at":"2026-07-05T11:30:52.806489+00:00"},{"alias_kind":"pith_short_8","alias_value":"F5FWAOGQ","created_at":"2026-07-05T11:30:52.806489+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":22,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.26997","citing_title":"RolloutPipe: Overlapping Pipelined Rollout and Training in Disaggregated On-Policy LLM Reinforcement Learning","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2607.01120","citing_title":"Next-Generation Agentic Reinforcement Learning Systems Enable Self-Evolving Agents","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2606.11867","citing_title":"Harnessing Routing Foresight for Micro-step-level MoE load balancing in RL Post-training","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2606.31410","citing_title":"Xiaomi-GUI-0 Technical Report","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2606.05597","citing_title":"AsyncWebRL: Efficient Multi-Step RL for Visual Web Agents","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2606.04560","citing_title":"Rollout-Level Advantage-Prioritized Experience Replay for GRPO","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2607.01120","citing_title":"Next-Generation Agentic Reinforcement Learning Systems Enable Self-Evolving Agents","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03077","citing_title":"Libra: Efficient Resource Management for Agentic RL Post-Training","ref_index":21,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13779","citing_title":"MinT: Managed Infrastructure for Training and Serving Millions of LLMs","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2606.31410","citing_title":"Xiaomi-GUI-0 Technical Report","ref_index":13,"is_internal_anchor":false},{"citing_arxiv_id":"2605.30859","citing_title":"DARTS: Distribution-Aware Active Rollout Trajectory Shaping for Accelerating LLM Reinforcement Learning","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06534","citing_title":"ROSE: Rollout On Serving GPUs via Cooperative Elasticity for Agentic RL","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2605.18799","citing_title":"ReCrit: Transition-Aware Reinforcement Learning for Scientific Critic Reasoning","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2605.15565","citing_title":"AstraFlow: Dataflow-Oriented Reinforcement Learning for Agentic LLMs","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2605.17570","citing_title":"How Off-Policy Can GRPO Be? Mu-GRPO for Efficient LLM Reinforcement Learning","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2510.19225","citing_title":"RLBoost: Harvesting Preemptible Resources for Cost-Efficient Reinforcement Learning on LLMs","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2511.14617","citing_title":"Seer: Online Context Learning for Fast Synchronous LLM Reinforcement Learning","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2512.12476","citing_title":"HetRL: Efficient Reinforcement Learning for LLMs in Heterogeneous Environments","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2605.13779","citing_title":"MinT: Managed Infrastructure for Training and Serving Millions of LLMs","ref_index":2,"is_internal_anchor":false},{"citing_arxiv_id":"2604.26256","citing_title":"DORA: A Scalable Asynchronous Reinforcement Learning System for Language Model Training","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06534","citing_title":"ROSE: Rollout On Serving GPUs via Cooperative Elasticity for Agentic RL","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2604.11554","citing_title":"Relax: An Asynchronous Reinforcement Learning Engine for Omni-Modal Post-Training at Scale","ref_index":9,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/F5FWAOGQMOMNKWULB3ENVPAJIZ","json":"https://pith.science/pith/F5FWAOGQMOMNKWULB3ENVPAJIZ.json","graph_json":"https://pith.science/api/pith-number/F5FWAOGQMOMNKWULB3ENVPAJIZ/graph.json","events_json":"https://pith.science/api/pith-number/F5FWAOGQMOMNKWULB3ENVPAJIZ/events.json","paper":"https://pith.science/paper/F5FWAOGQ"},"agent_actions":{"view_html":"https://pith.science/pith/F5FWAOGQMOMNKWULB3ENVPAJIZ","download_json":"https://pith.science/pith/F5FWAOGQMOMNKWULB3ENVPAJIZ.json","view_paper":"https://pith.science/paper/F5FWAOGQ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.01663&json=true","fetch_graph":"https://pith.science/api/pith-number/F5FWAOGQMOMNKWULB3ENVPAJIZ/graph.json","fetch_events":"https://pith.science/api/pith-number/F5FWAOGQMOMNKWULB3ENVPAJIZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/F5FWAOGQMOMNKWULB3ENVPAJIZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/F5FWAOGQMOMNKWULB3ENVPAJIZ/action/storage_attestation","attest_author":"https://pith.science/pith/F5FWAOGQMOMNKWULB3ENVPAJIZ/action/author_attestation","sign_citation":"https://pith.science/pith/F5FWAOGQMOMNKWULB3ENVPAJIZ/action/citation_signature","submit_replication":"https://pith.science/pith/F5FWAOGQMOMNKWULB3ENVPAJIZ/action/replication_record"}},"created_at":"2026-07-05T11:30:52.806489+00:00","updated_at":"2026-07-05T11:30:52.806489+00:00"}