{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:2DDT7GQNFROWPUW73U6X7Y5INL","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"62d9efd6af57211c0680900539c73ebdea02a5deb0ebf6f19c97d75a22120c32","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T16:14:46Z","title_canon_sha256":"0845ff7196936e93f23f3c82f8bbcc355bf9f08909a3858c46573b9b7ca8c249"},"schema_version":"1.0","source":{"id":"2605.13935","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13935","created_at":"2026-05-17T23:39:13Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13935v1","created_at":"2026-05-17T23:39:13Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13935","created_at":"2026-05-17T23:39:13Z"},{"alias_kind":"pith_short_12","alias_value":"2DDT7GQNFROW","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"2DDT7GQNFROWPUW7","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"2DDT7GQN","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:056f8fa18ba1380f78ba7d5d5768049395ff5864e0475e9e37635d6affcaf07d","target":"graph","created_at":"2026-05-17T23:39:13Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"TraFL is the only evaluated post-training method that improves over the base model in every benchmark-length setting, with gains that persist as the sampling budget increases."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The diffusion-compatible sequence-level surrogate and learned prompt-dependent normalization faithfully approximate the trajectory-balance objective without introducing new collapse modes or requiring task-specific tuning."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"TraFL applies trajectory flow balancing to post-train diffusion language models, preventing mode collapse and delivering consistent gains on reasoning tasks that hold under increased sampling."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"A trajectory-balance objective stops diffusion language models from locking onto narrow denoising paths during post-training."}],"snapshot_sha256":"8e66d9b26f83d2f5f8ad7422e648b699deaff32db4069de169cb99ac3e95abec"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"75089fe02f257850b4c229472635656a8527d8e1c2889c4207a565211692b69d"},"paper":{"abstract_excerpt":"Diffusion language models are a promising alternative to autoregressive models, yet post-training methods for them largely adapt reward-maximizing objectives. We identify a central failure mode in this setting we call trajectory locking: sampled reward-driven updates over-concentrate probability mass onto a narrow set of denoising paths, reducing coverage of alternative correct solutions under repeated sampling. To address this, we propose TraFL (Trajectory Flow baLancing), a trajectory-balance objective that trains the policy toward a reward-tilted target distribution anchored to a frozen ref","authors_text":"Prasanna Parthasarathi, Saba Ahmadi, Yufei Cui","cross_cats":["cs.CL"],"headline":"A trajectory-balance objective stops diffusion language models from locking onto narrow denoising paths during post-training.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T16:14:46Z","title":"Beyond Mode-Seeking RL: Trajectory-Balance Post-Training for Diffusion Language Models"},"references":{"count":32,"internal_anchors":7,"resolved_work":32,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Natural gradient works efficiently in learning.Neural computation, 10(2):251– 276, 1998","work_id":"b9a8c272-145a-4fe5-930b-43b05b08fcd2","year":1998},{"cited_arxiv_id":"2108.07732","doi":"","is_internal_anchor":true,"ref_index":2,"title":"Program Synthesis with Large Language Models","work_id":"fd241a05-03b9-4de2-9588-9d77ce176125","year":2021},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Mirror descent and nonlinear projected subgradient methods for convex optimization.Operations Research Letters, 31(3):167–175","work_id":"d83325bc-fa70-4661-b21c-a8416937436b","year":2003},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Hu, Mo Tiwari, and Emmanuel Bengio","work_id":"185d0c88-9be3-4617-a532-40e368019cfc","year":2023},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde de Oliveira Pinto, Jared Kaplan, Harri Edwards, Yuri Burda, Nicholas Joseph, Greg Brockman, Alex Ray, Raul Puri, Gretchen Krueger, Mich","work_id":"fbf5883a-dc87-439f-a0ae-fcc99bbb1643","year":2021}],"snapshot_sha256":"8812a7d525a5777a87bb5b86aff9793e63b2b78593e41e614a6fdb8a2b9f3a7c"},"source":{"id":"2605.13935","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-15T04:56:08.731453Z","id":"e6e37a21-437f-4a2f-83f4-c3f8b182175b","model_set":{"reader":"grok-4.3"},"one_line_summary":"TraFL applies trajectory flow balancing to post-train diffusion language models, preventing mode collapse and delivering consistent gains on reasoning tasks that hold under increased sampling.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"A trajectory-balance objective stops diffusion language models from locking onto narrow denoising paths during post-training.","strongest_claim":"TraFL is the only evaluated post-training method that improves over the base model in every benchmark-length setting, with gains that persist as the sampling budget increases.","weakest_assumption":"The diffusion-compatible sequence-level surrogate and learned prompt-dependent normalization faithfully approximate the trajectory-balance objective without introducing new collapse modes or requiring task-specific tuning."}},"verdict_id":"e6e37a21-437f-4a2f-83f4-c3f8b182175b"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:f8d45e4bc1538ec06c57e98e4a27479d8c8073d94cad5c2ae5715f453846d41d","target":"record","created_at":"2026-05-17T23:39:13Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"62d9efd6af57211c0680900539c73ebdea02a5deb0ebf6f19c97d75a22120c32","cross_cats_sorted":["cs.CL"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-13T16:14:46Z","title_canon_sha256":"0845ff7196936e93f23f3c82f8bbcc355bf9f08909a3858c46573b9b7ca8c249"},"schema_version":"1.0","source":{"id":"2605.13935","kind":"arxiv","version":1}},"canonical_sha256":"d0c73f9a0d2c5d67d2dfdd3d7fe3a86af9d4b1b9f5f46ece921496f831f4370f","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"d0c73f9a0d2c5d67d2dfdd3d7fe3a86af9d4b1b9f5f46ece921496f831f4370f","first_computed_at":"2026-05-17T23:39:13.930271Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:39:13.930271Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"H+N55xp2aOcaG27CNeuTRp0HH7u3MH61IYPTUYbAlZa2uaNLAdGWG2FXNKuMJHayx1o9eohWFsElOND6RYxhCA==","signature_status":"signed_v1","signed_at":"2026-05-17T23:39:13.931038Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.13935","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:f8d45e4bc1538ec06c57e98e4a27479d8c8073d94cad5c2ae5715f453846d41d","sha256:056f8fa18ba1380f78ba7d5d5768049395ff5864e0475e9e37635d6affcaf07d"],"state_sha256":"b4d5e7e5914835336af284364a29fbb4400ca84d0abd2a81c662f027ce712579"}