{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:C2JAM3NK5YGL7RX6FLMB73WJOI","short_pith_number":"pith:C2JAM3NK","canonical_record":{"source":{"id":"2604.03118","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-04-03T15:43:21Z","cross_cats_sorted":["eess.IV"],"title_canon_sha256":"1f81f90fc8f84eee1cad6c816ad45b6b5266f180f735a92289f264f474f1ece6","abstract_canon_sha256":"2e5b42066cdece165e571d2c17fce0e5582fb712271a3342f2469b7e8aa7f87c"},"schema_version":"1.0"},"canonical_sha256":"1692066daaee0cbfc6fe2ad81feec9723902bb4d9e73a88769cb1904398fc75b","source":{"kind":"arxiv","id":"2604.03118","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2604.03118","created_at":"2026-07-02T01:17:30Z"},{"alias_kind":"arxiv_version","alias_value":"2604.03118v2","created_at":"2026-07-02T01:17:30Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.03118","created_at":"2026-07-02T01:17:30Z"},{"alias_kind":"pith_short_12","alias_value":"C2JAM3NK5YGL","created_at":"2026-07-02T01:17:30Z"},{"alias_kind":"pith_short_16","alias_value":"C2JAM3NK5YGL7RX6","created_at":"2026-07-02T01:17:30Z"},{"alias_kind":"pith_short_8","alias_value":"C2JAM3NK","created_at":"2026-07-02T01:17:30Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:C2JAM3NK5YGL7RX6FLMB73WJOI","target":"record","payload":{"canonical_record":{"source":{"id":"2604.03118","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-04-03T15:43:21Z","cross_cats_sorted":["eess.IV"],"title_canon_sha256":"1f81f90fc8f84eee1cad6c816ad45b6b5266f180f735a92289f264f474f1ece6","abstract_canon_sha256":"2e5b42066cdece165e571d2c17fce0e5582fb712271a3342f2469b7e8aa7f87c"},"schema_version":"1.0"},"canonical_sha256":"1692066daaee0cbfc6fe2ad81feec9723902bb4d9e73a88769cb1904398fc75b","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-02T01:17:30.449463Z","signature_b64":"EWWobVA2m3EfDPBiCfg6yxFl/k8siX+kKv2NfWYdGzvbYd1mcVNNgLwDr/VSHMab601q1MF49DNhDHiSEOBRDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1692066daaee0cbfc6fe2ad81feec9723902bb4d9e73a88769cb1904398fc75b","last_reissued_at":"2026-07-02T01:17:30.448942Z","signature_status":"signed_v1","first_computed_at":"2026-07-02T01:17:30.448942Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2604.03118","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-02T01:17:30Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"ILubuYput27noPmsBQZkG1zRSHIkeHDgjUOmUXk9A1PUp7Gb0Lhx57ld02esjVhqvwWCXa4PWoEZesrZ5TpMDA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-05T12:19:42.005543Z"},"content_sha256":"9a03236f8fb4be2a15661f00ca7bbdd6d7576c5fb12d93c5a0b0bcd66e12ed88","schema_version":"1.0","event_id":"sha256:9a03236f8fb4be2a15661f00ca7bbdd6d7576c5fb12d93c5a0b0bcd66e12ed88"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:C2JAM3NK5YGL7RX6FLMB73WJOI","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Salt: Self-Consistent Distribution Matching with Cache-Aware Training for Fast Video Generation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Salt distills video models to 2-4 steps by regularizing the endpoint consistency of consecutive denoising updates and conditioning on KV cache states.","cross_cats":["eess.IV"],"primary_cat":"cs.CV","authors_text":"Bingqi Ma, Dailan He, Guanglu Song, Jun Zhang, Xiahong Wang, Xingtong Ge, Yi Zhang, Yu Liu, Yushi Huang","submitted_at":"2026-04-03T15:43:21Z","abstract_excerpt":"Distilling video generation models to extremely low inference budgets (e.g., 2--4 NFEs) is crucial for real-time deployment, yet remains challenging. Trajectory-style consistency distillation often becomes conservative under complex video dynamics, yielding an over-smoothed appearance and weak motion. Distribution matching distillation (DMD) can recover sharp, mode-seeking samples, but its local training signals do not explicitly regularize how denoising updates compose across timesteps, making composed rollouts prone to drift. To overcome this challenge, we propose Self-Consistent Distributio"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Across extensive experiments on both non-autoregressive backbones (e.g., Wan 2.1) and autoregressive real-time paradigms (e.g., Self Forcing), our method, dubbed Salt, consistently improves low-NFE video generation quality while remaining compatible with diverse KV-cache memory mechanisms.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That explicitly regularizing the endpoint-consistent composition of consecutive denoising updates will prevent drift in composed rollouts, and that the cache-conditioned feature alignment will reliably steer low-quality outputs toward high-quality references without introducing new inconsistencies.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Salt improves low-step video generation quality by adding endpoint-consistent regularization to distribution matching distillation and using cache-conditioned feature alignment for autoregressive models.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Salt distills video models to 2-4 steps by regularizing the endpoint consistency of consecutive denoising updates and conditioning on KV cache states.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"e61415009b9903b5b935814bcc2518e5c3e72ee3209f655ba4950ae499913a52"},"source":{"id":"2604.03118","kind":"arxiv","version":2},"verdict":{"id":"7e06cb90-3026-4b27-bdb4-05cf9ae4fe50","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-13T19:35:26.286285Z","strongest_claim":"Across extensive experiments on both non-autoregressive backbones (e.g., Wan 2.1) and autoregressive real-time paradigms (e.g., Self Forcing), our method, dubbed Salt, consistently improves low-NFE video generation quality while remaining compatible with diverse KV-cache memory mechanisms.","one_line_summary":"Salt improves low-step video generation quality by adding endpoint-consistent regularization to distribution matching distillation and using cache-conditioned feature alignment for autoregressive models.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That explicitly regularizing the endpoint-consistent composition of consecutive denoising updates will prevent drift in composed rollouts, and that the cache-conditioned feature alignment will reliably steer low-quality outputs toward high-quality references without introducing new inconsistencies.","pith_extraction_headline":"Salt distills video models to 2-4 steps by regularizing the endpoint consistency of consecutive denoising updates and conditioning on KV cache states."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.03118/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":2,"snapshot_sha256":"8ec0f5c68f0b1507d7aebce363ae127ab1481e81b2fb6d769b7c3be013fb19c0"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"7e06cb90-3026-4b27-bdb4-05cf9ae4fe50"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-02T01:17:30Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"4Kcq2vKILQFYBeeYeVsDqizp0O3ZxODS76rJwrPorEXAvANGgvD4YQLCKyFlqrjaeTtSOq9EuX2nLfv3OihXDA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-05T12:19:42.006018Z"},"content_sha256":"c7afcd7654b8c0c2b4b84a0dfdfdb980b706d9afdffa772ce4ce73e26da4d0f9","schema_version":"1.0","event_id":"sha256:c7afcd7654b8c0c2b4b84a0dfdfdb980b706d9afdffa772ce4ce73e26da4d0f9"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/C2JAM3NK5YGL7RX6FLMB73WJOI/bundle.json","state_url":"https://pith.science/pith/C2JAM3NK5YGL7RX6FLMB73WJOI/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/C2JAM3NK5YGL7RX6FLMB73WJOI/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-07-05T12:19:42Z","links":{"resolver":"https://pith.science/pith/C2JAM3NK5YGL7RX6FLMB73WJOI","bundle":"https://pith.science/pith/C2JAM3NK5YGL7RX6FLMB73WJOI/bundle.json","state":"https://pith.science/pith/C2JAM3NK5YGL7RX6FLMB73WJOI/state.json","well_known_bundle":"https://pith.science/.well-known/pith/C2JAM3NK5YGL7RX6FLMB73WJOI/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:C2JAM3NK5YGL7RX6FLMB73WJOI","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":"2e5b42066cdece165e571d2c17fce0e5582fb712271a3342f2469b7e8aa7f87c","cross_cats_sorted":["eess.IV"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-04-03T15:43:21Z","title_canon_sha256":"1f81f90fc8f84eee1cad6c816ad45b6b5266f180f735a92289f264f474f1ece6"},"schema_version":"1.0","source":{"id":"2604.03118","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2604.03118","created_at":"2026-07-02T01:17:30Z"},{"alias_kind":"arxiv_version","alias_value":"2604.03118v2","created_at":"2026-07-02T01:17:30Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.03118","created_at":"2026-07-02T01:17:30Z"},{"alias_kind":"pith_short_12","alias_value":"C2JAM3NK5YGL","created_at":"2026-07-02T01:17:30Z"},{"alias_kind":"pith_short_16","alias_value":"C2JAM3NK5YGL7RX6","created_at":"2026-07-02T01:17:30Z"},{"alias_kind":"pith_short_8","alias_value":"C2JAM3NK","created_at":"2026-07-02T01:17:30Z"}],"graph_snapshots":[{"event_id":"sha256:c7afcd7654b8c0c2b4b84a0dfdfdb980b706d9afdffa772ce4ce73e26da4d0f9","target":"graph","created_at":"2026-07-02T01:17:30Z","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":"Across extensive experiments on both non-autoregressive backbones (e.g., Wan 2.1) and autoregressive real-time paradigms (e.g., Self Forcing), our method, dubbed Salt, consistently improves low-NFE video generation quality while remaining compatible with diverse KV-cache memory mechanisms."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That explicitly regularizing the endpoint-consistent composition of consecutive denoising updates will prevent drift in composed rollouts, and that the cache-conditioned feature alignment will reliably steer low-quality outputs toward high-quality references without introducing new inconsistencies."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Salt improves low-step video generation quality by adding endpoint-consistent regularization to distribution matching distillation and using cache-conditioned feature alignment for autoregressive models."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Salt distills video models to 2-4 steps by regularizing the endpoint consistency of consecutive denoising updates and conditioning on KV cache states."}],"snapshot_sha256":"e61415009b9903b5b935814bcc2518e5c3e72ee3209f655ba4950ae499913a52"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"8ec0f5c68f0b1507d7aebce363ae127ab1481e81b2fb6d769b7c3be013fb19c0"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2604.03118/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Distilling video generation models to extremely low inference budgets (e.g., 2--4 NFEs) is crucial for real-time deployment, yet remains challenging. Trajectory-style consistency distillation often becomes conservative under complex video dynamics, yielding an over-smoothed appearance and weak motion. Distribution matching distillation (DMD) can recover sharp, mode-seeking samples, but its local training signals do not explicitly regularize how denoising updates compose across timesteps, making composed rollouts prone to drift. To overcome this challenge, we propose Self-Consistent Distributio","authors_text":"Bingqi Ma, Dailan He, Guanglu Song, Jun Zhang, Xiahong Wang, Xingtong Ge, Yi Zhang, Yu Liu, Yushi Huang","cross_cats":["eess.IV"],"headline":"Salt distills video models to 2-4 steps by regularizing the endpoint consistency of consecutive denoising updates and conditioning on KV cache states.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-04-03T15:43:21Z","title":"Salt: Self-Consistent Distribution Matching with Cache-Aware Training for Fast Video Generation"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2604.03118","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-13T19:35:26.286285Z","id":"7e06cb90-3026-4b27-bdb4-05cf9ae4fe50","model_set":{"reader":"grok-4.3"},"one_line_summary":"Salt improves low-step video generation quality by adding endpoint-consistent regularization to distribution matching distillation and using cache-conditioned feature alignment for autoregressive models.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Salt distills video models to 2-4 steps by regularizing the endpoint consistency of consecutive denoising updates and conditioning on KV cache states.","strongest_claim":"Across extensive experiments on both non-autoregressive backbones (e.g., Wan 2.1) and autoregressive real-time paradigms (e.g., Self Forcing), our method, dubbed Salt, consistently improves low-NFE video generation quality while remaining compatible with diverse KV-cache memory mechanisms.","weakest_assumption":"That explicitly regularizing the endpoint-consistent composition of consecutive denoising updates will prevent drift in composed rollouts, and that the cache-conditioned feature alignment will reliably steer low-quality outputs toward high-quality references without introducing new inconsistencies."}},"verdict_id":"7e06cb90-3026-4b27-bdb4-05cf9ae4fe50"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:9a03236f8fb4be2a15661f00ca7bbdd6d7576c5fb12d93c5a0b0bcd66e12ed88","target":"record","created_at":"2026-07-02T01:17:30Z","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":"2e5b42066cdece165e571d2c17fce0e5582fb712271a3342f2469b7e8aa7f87c","cross_cats_sorted":["eess.IV"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-04-03T15:43:21Z","title_canon_sha256":"1f81f90fc8f84eee1cad6c816ad45b6b5266f180f735a92289f264f474f1ece6"},"schema_version":"1.0","source":{"id":"2604.03118","kind":"arxiv","version":2}},"canonical_sha256":"1692066daaee0cbfc6fe2ad81feec9723902bb4d9e73a88769cb1904398fc75b","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"1692066daaee0cbfc6fe2ad81feec9723902bb4d9e73a88769cb1904398fc75b","first_computed_at":"2026-07-02T01:17:30.448942Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-02T01:17:30.448942Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"EWWobVA2m3EfDPBiCfg6yxFl/k8siX+kKv2NfWYdGzvbYd1mcVNNgLwDr/VSHMab601q1MF49DNhDHiSEOBRDA==","signature_status":"signed_v1","signed_at":"2026-07-02T01:17:30.449463Z","signed_message":"canonical_sha256_bytes"},"source_id":"2604.03118","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:9a03236f8fb4be2a15661f00ca7bbdd6d7576c5fb12d93c5a0b0bcd66e12ed88","sha256:c7afcd7654b8c0c2b4b84a0dfdfdb980b706d9afdffa772ce4ce73e26da4d0f9"],"state_sha256":"154228f61e447a7eac8028df0f4e5e2a4c951e72f576680faae5c14aa17d6a45"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"N2lvZ4iDjFHg/Sa3Sp+MBGpzJmD5QQLfNKxrzuLiDt0fnXh/tdG4mMdtuGLrM7AHeJQwg5wq9Q3Eq4J5tZGmCA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-07-05T12:19:42.008492Z","bundle_sha256":"65b65294f0665c1db53ca8160e11c096867563f169b951beda61978d14da0209"}}