{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:MKVLOMPPP2RBPXQACF4YPBHM2U","short_pith_number":"pith:MKVLOMPP","schema_version":"1.0","canonical_sha256":"62aab731ef7ea217de0011798784ecd52cdd7a6e318671dc438dda2899051239","source":{"kind":"arxiv","id":"2508.04324","version":4},"attestation_state":"computed","paper":{"title":"TempFlow-GRPO: When Timing Matters for GRPO in Flow Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Bo Zhang, Dacheng Yin, Fengyun Rao, Jian Yang, Siming Fu, Wanli Li, Xiaoxuan He, Yuke Zhao","submitted_at":"2025-08-06T11:10:39Z","abstract_excerpt":"Recent flow matching models for text-to-image generation have achieved remarkable quality, yet their integration with reinforcement learning for human preference alignment remains suboptimal, hindering fine-grained reward-based optimization. We observe that the key impediment to effective GRPO training of flow models is the temporal uniformity assumption in existing approaches: sparse terminal rewards with uniform credit assignment fail to capture the varying criticality of decisions across generation timesteps, resulting in inefficient exploration and suboptimal convergence. To remedy this sh"},"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":"2508.04324","kind":"arxiv","version":4},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2025-08-06T11:10:39Z","cross_cats_sorted":[],"title_canon_sha256":"6de3a96109dfeac090c1bfdc441f39f1288a4549c8d44e9b0d50f438c07d94c8","abstract_canon_sha256":"7c8d6a920e5527a95130b88e1512ceeb717cd047c2ec00aa2e65a98a0236ed89"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-21T21:45:38.093040Z","signature_b64":"axhPsVERPwD6NVk+UMOcDR47HVYPqeWlN/zS8ZjdIbKKRW+yuxb7Xd3JNCL+DMYjbN9obuS/zEPOsSvHFOQeAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"62aab731ef7ea217de0011798784ecd52cdd7a6e318671dc438dda2899051239","last_reissued_at":"2026-05-21T21:45:38.091232Z","signature_status":"signed_v1","first_computed_at":"2026-05-21T21:45:38.091232Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"TempFlow-GRPO: When Timing Matters for GRPO in Flow Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Bo Zhang, Dacheng Yin, Fengyun Rao, Jian Yang, Siming Fu, Wanli Li, Xiaoxuan He, Yuke Zhao","submitted_at":"2025-08-06T11:10:39Z","abstract_excerpt":"Recent flow matching models for text-to-image generation have achieved remarkable quality, yet their integration with reinforcement learning for human preference alignment remains suboptimal, hindering fine-grained reward-based optimization. We observe that the key impediment to effective GRPO training of flow models is the temporal uniformity assumption in existing approaches: sparse terminal rewards with uniform credit assignment fail to capture the varying criticality of decisions across generation timesteps, resulting in inefficient exploration and suboptimal convergence. To remedy this sh"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.04324","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/2508.04324/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":"2508.04324","created_at":"2026-05-21T21:45:38.091330+00:00"},{"alias_kind":"arxiv_version","alias_value":"2508.04324v4","created_at":"2026-05-21T21:45:38.091330+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.04324","created_at":"2026-05-21T21:45:38.091330+00:00"},{"alias_kind":"pith_short_12","alias_value":"MKVLOMPPP2RB","created_at":"2026-05-21T21:45:38.091330+00:00"},{"alias_kind":"pith_short_16","alias_value":"MKVLOMPPP2RBPXQA","created_at":"2026-05-21T21:45:38.091330+00:00"},{"alias_kind":"pith_short_8","alias_value":"MKVLOMPP","created_at":"2026-05-21T21:45:38.091330+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":21,"internal_anchor_count":21,"sample":[{"citing_arxiv_id":"2509.23352","citing_title":"Dynamic-TreeRPO: Breaking the Independent Trajectory Bottleneck with Structured Sampling","ref_index":5,"is_internal_anchor":true},{"citing_arxiv_id":"2510.21583","citing_title":"Principled RL for Flow Matching Emerges from the Chunk-level Policy Optimization","ref_index":3,"is_internal_anchor":true},{"citing_arxiv_id":"2511.18719","citing_title":"Seeing What Matters: Visual Preference Policy Optimization for Visual Generation","ref_index":9,"is_internal_anchor":true},{"citing_arxiv_id":"2602.03139","citing_title":"Diversity-Preserved Distribution Matching Distillation for Fast Visual Synthesis","ref_index":2,"is_internal_anchor":true},{"citing_arxiv_id":"2602.04663","citing_title":"Rethinking the Design Space of Reinforcement Learning for Diffusion Models: On the Importance of Likelihood Estimation Beyond Loss Design","ref_index":7,"is_internal_anchor":true},{"citing_arxiv_id":"2605.15803","citing_title":"Embedding-perturbed Exploration Preference Optimization for Flow Models","ref_index":35,"is_internal_anchor":true},{"citing_arxiv_id":"2605.15980","citing_title":"Flash-GRPO: Efficient Alignment for Video Diffusion via One-Step Policy Optimization","ref_index":5,"is_internal_anchor":true},{"citing_arxiv_id":"2508.20751","citing_title":"Pref-GRPO: Pairwise Preference Reward-based GRPO for Stable Text-to-Image Reinforcement Learning","ref_index":7,"is_internal_anchor":true},{"citing_arxiv_id":"2509.08827","citing_title":"A Survey of Reinforcement Learning for Large Reasoning Models","ref_index":192,"is_internal_anchor":true},{"citing_arxiv_id":"2605.12480","citing_title":"OmniNFT: Modality-wise Omni Diffusion Reinforcement for Joint Audio-Video Generation","ref_index":7,"is_internal_anchor":true},{"citing_arxiv_id":"2605.12112","citing_title":"When Policy Entropy Constraint Fails: Preserving Diversity in Flow-based RLHF via Perceptual Entropy","ref_index":26,"is_internal_anchor":true},{"citing_arxiv_id":"2605.10937","citing_title":"Power Reinforcement Post-Training of Text-to-Image Models with Super-Linear Advantage Shaping","ref_index":99,"is_internal_anchor":true},{"citing_arxiv_id":"2604.25427","citing_title":"A Systematic Post-Train Framework for Video Generation","ref_index":25,"is_internal_anchor":true},{"citing_arxiv_id":"2604.23380","citing_title":"V-GRPO: Online Reinforcement Learning for Denoising Generative Models Is Easier than You Think","ref_index":8,"is_internal_anchor":true},{"citing_arxiv_id":"2604.19234","citing_title":"Learning to Credit the Right Steps: Objective-aware Process Optimization for Visual Generation","ref_index":14,"is_internal_anchor":true},{"citing_arxiv_id":"2604.06966","citing_title":"MAR-GRPO: Stabilized GRPO for AR-diffusion Hybrid Image Generation","ref_index":11,"is_internal_anchor":true},{"citing_arxiv_id":"2604.06916","citing_title":"FP4 Explore, BF16 Train: Diffusion Reinforcement Learning via Efficient Rollout Scaling","ref_index":20,"is_internal_anchor":true},{"citing_arxiv_id":"2604.14910","citing_title":"Reward-Aware Trajectory Shaping for Few-step Visual Generation","ref_index":8,"is_internal_anchor":true},{"citing_arxiv_id":"2604.18518","citing_title":"UDM-GRPO: Stable and Efficient Group Relative Policy Optimization for Uniform Discrete Diffusion Models","ref_index":6,"is_internal_anchor":true},{"citing_arxiv_id":"2604.19406","citing_title":"HP-Edit: A Human-Preference Post-Training Framework for Image Editing","ref_index":11,"is_internal_anchor":true},{"citing_arxiv_id":"2604.19009","citing_title":"Guiding Distribution Matching Distillation with Gradient-Based Reinforcement Learning","ref_index":12,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/MKVLOMPPP2RBPXQACF4YPBHM2U","json":"https://pith.science/pith/MKVLOMPPP2RBPXQACF4YPBHM2U.json","graph_json":"https://pith.science/api/pith-number/MKVLOMPPP2RBPXQACF4YPBHM2U/graph.json","events_json":"https://pith.science/api/pith-number/MKVLOMPPP2RBPXQACF4YPBHM2U/events.json","paper":"https://pith.science/paper/MKVLOMPP"},"agent_actions":{"view_html":"https://pith.science/pith/MKVLOMPPP2RBPXQACF4YPBHM2U","download_json":"https://pith.science/pith/MKVLOMPPP2RBPXQACF4YPBHM2U.json","view_paper":"https://pith.science/paper/MKVLOMPP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2508.04324&json=true","fetch_graph":"https://pith.science/api/pith-number/MKVLOMPPP2RBPXQACF4YPBHM2U/graph.json","fetch_events":"https://pith.science/api/pith-number/MKVLOMPPP2RBPXQACF4YPBHM2U/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/MKVLOMPPP2RBPXQACF4YPBHM2U/action/timestamp_anchor","attest_storage":"https://pith.science/pith/MKVLOMPPP2RBPXQACF4YPBHM2U/action/storage_attestation","attest_author":"https://pith.science/pith/MKVLOMPPP2RBPXQACF4YPBHM2U/action/author_attestation","sign_citation":"https://pith.science/pith/MKVLOMPPP2RBPXQACF4YPBHM2U/action/citation_signature","submit_replication":"https://pith.science/pith/MKVLOMPPP2RBPXQACF4YPBHM2U/action/replication_record"}},"created_at":"2026-05-21T21:45:38.091330+00:00","updated_at":"2026-05-21T21:45:38.091330+00:00"}