{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:4TT7PDU2ST3K2TTCID2DXH3T5H","short_pith_number":"pith:4TT7PDU2","canonical_record":{"source":{"id":"2605.02772","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-05-04T16:16:26Z","cross_cats_sorted":[],"title_canon_sha256":"b71a814c742e400a2f004d83bceca10675a1a765736b26042028c90b8cdbaa39","abstract_canon_sha256":"c70e9ae07c03a95926855e51fc82a37964e7617dc3da0f0f41fa056adff106f7"},"schema_version":"1.0"},"canonical_sha256":"e4e7f78e9a94f6ad4e6240f43b9f73e9e0fba53aaa694895847f3ea0ef93094b","source":{"kind":"arxiv","id":"2605.02772","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.02772","created_at":"2026-05-29T01:05:11Z"},{"alias_kind":"arxiv_version","alias_value":"2605.02772v2","created_at":"2026-05-29T01:05:11Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.02772","created_at":"2026-05-29T01:05:11Z"},{"alias_kind":"pith_short_12","alias_value":"4TT7PDU2ST3K","created_at":"2026-05-29T01:05:11Z"},{"alias_kind":"pith_short_16","alias_value":"4TT7PDU2ST3K2TTC","created_at":"2026-05-29T01:05:11Z"},{"alias_kind":"pith_short_8","alias_value":"4TT7PDU2","created_at":"2026-05-29T01:05:11Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:4TT7PDU2ST3K2TTCID2DXH3T5H","target":"record","payload":{"canonical_record":{"source":{"id":"2605.02772","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-05-04T16:16:26Z","cross_cats_sorted":[],"title_canon_sha256":"b71a814c742e400a2f004d83bceca10675a1a765736b26042028c90b8cdbaa39","abstract_canon_sha256":"c70e9ae07c03a95926855e51fc82a37964e7617dc3da0f0f41fa056adff106f7"},"schema_version":"1.0"},"canonical_sha256":"e4e7f78e9a94f6ad4e6240f43b9f73e9e0fba53aaa694895847f3ea0ef93094b","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-29T01:05:11.530905Z","signature_b64":"4ZhQi+/R1n0PY69wZE9+ogDCcmxic10bwjjg0D1W8+aWR4tlM+xzbc8DVP08KGPvC+/RBdQ+uGmJ1dVERUsEAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e4e7f78e9a94f6ad4e6240f43b9f73e9e0fba53aaa694895847f3ea0ef93094b","last_reissued_at":"2026-05-29T01:05:11.529602Z","signature_status":"signed_v1","first_computed_at":"2026-05-29T01:05:11.529602Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.02772","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":"2026-05-29T01:05:11Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"LYmCknCu+074RhD66nFOQENk3EZGxLj3RI2MmbUpo7GoSeojqVSMXURdD2RJCLbRuZkJlJEnV5DylX7a2WXwBQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-08T10:40:02.769948Z"},"content_sha256":"f8cec3a1fcd99f93aa97b1178ba63e5ba78776beb0b95e4f4f1f6875f996b525","schema_version":"1.0","event_id":"sha256:f8cec3a1fcd99f93aa97b1178ba63e5ba78776beb0b95e4f4f1f6875f996b525"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:4TT7PDU2ST3K2TTCID2DXH3T5H","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Linearizing Vision Transformer with Test-Time Training","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Test-Time Training aligns linear attention with pretrained Softmax weights, enabling transfer after minimal fine-tuning.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Dongchen Han, Gao Huang, Hanyi Wang, Yining Li, Yulin Wang, Zeyu Liu","submitted_at":"2026-05-04T16:16:26Z","abstract_excerpt":"While linear-complexity attention mechanisms offer a promising alternative to Softmax attention for overcoming the quadratic bottleneck, training such models from scratch remains prohibitively expensive. Inheriting weights from pretrained Transformers provides an appealing shortcut, yet the fundamental representational gap between Softmax and linear attention prevents effective weight transfer. In this work, we address this conversion challenge from two perspectives: architectural alignment and representational alignment. We identify Test-Time Training (TTT) as a linear-complexity architecture"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"With only 1 hour of fine-tuning on 4×H20 GPUs, SD3.5-T^5 achieves comparable text-to-image quality to the fine-tuned Softmax model, while accelerating inference by 1.32× and 1.47× at 1K and 2K resolutions.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The representational gap between Softmax and linear attention can be closed sufficiently by TTT's two-layer dynamic formulation plus the introduced key instance normalization and lightweight locality enhancement module to allow effective weight inheritance.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Using Test-Time Training's structural match to Softmax attention plus key normalization and locality modules allows inheriting pretrained weights and fine-tuning Stable Diffusion 3.5 in one hour to match quality while speeding inference 1.32-1.47x.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Test-Time Training aligns linear attention with pretrained Softmax weights, enabling transfer after minimal fine-tuning.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"93fc46772cf804700f368f4a611d938b6b078ac0e64a8364c7a315c6a6592174"},"source":{"id":"2605.02772","kind":"arxiv","version":2},"verdict":{"id":"87624763-74ea-45ed-9738-3fa1bb313768","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-08T18:22:17.491773Z","strongest_claim":"With only 1 hour of fine-tuning on 4×H20 GPUs, SD3.5-T^5 achieves comparable text-to-image quality to the fine-tuned Softmax model, while accelerating inference by 1.32× and 1.47× at 1K and 2K resolutions.","one_line_summary":"Using Test-Time Training's structural match to Softmax attention plus key normalization and locality modules allows inheriting pretrained weights and fine-tuning Stable Diffusion 3.5 in one hour to match quality while speeding inference 1.32-1.47x.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The representational gap between Softmax and linear attention can be closed sufficiently by TTT's two-layer dynamic formulation plus the introduced key instance normalization and lightweight locality enhancement module to allow effective weight inheritance.","pith_extraction_headline":"Test-Time Training aligns linear attention with pretrained Softmax weights, enabling transfer after minimal fine-tuning."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.02772/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"ai_meta_artifact","ran_at":"2026-05-20T15:34:05.169126Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-20T02:31:22.311632Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T15:59:30.513388Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"8a8ce11c3c63c50da39a97018aa53836631ea6d397cc20d8f1ba69798d81f2bb"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":3,"snapshot_sha256":"dc11f4d94d5680334403414253b47a6a791b65358a24294f407ef69b693e680a"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"87624763-74ea-45ed-9738-3fa1bb313768"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-29T01:05:11Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"0PYKn0ldlUmTzQ1Y+g+bgW3oXaH/XJVRyTfLDyfKAnWs9iGkfifADlQP6RTBRfXZuhKdwGXhx85B2Mae5s8PAg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-06-08T10:40:02.771009Z"},"content_sha256":"82d00275a1ad24415c3cd7c8a94336aa41a6c7c6c54c3e2e2f8d73d928cf1b67","schema_version":"1.0","event_id":"sha256:82d00275a1ad24415c3cd7c8a94336aa41a6c7c6c54c3e2e2f8d73d928cf1b67"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/4TT7PDU2ST3K2TTCID2DXH3T5H/bundle.json","state_url":"https://pith.science/pith/4TT7PDU2ST3K2TTCID2DXH3T5H/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/4TT7PDU2ST3K2TTCID2DXH3T5H/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":"2026-06-08T10:40:02Z","links":{"resolver":"https://pith.science/pith/4TT7PDU2ST3K2TTCID2DXH3T5H","bundle":"https://pith.science/pith/4TT7PDU2ST3K2TTCID2DXH3T5H/bundle.json","state":"https://pith.science/pith/4TT7PDU2ST3K2TTCID2DXH3T5H/state.json","well_known_bundle":"https://pith.science/.well-known/pith/4TT7PDU2ST3K2TTCID2DXH3T5H/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:4TT7PDU2ST3K2TTCID2DXH3T5H","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":"c70e9ae07c03a95926855e51fc82a37964e7617dc3da0f0f41fa056adff106f7","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-05-04T16:16:26Z","title_canon_sha256":"b71a814c742e400a2f004d83bceca10675a1a765736b26042028c90b8cdbaa39"},"schema_version":"1.0","source":{"id":"2605.02772","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.02772","created_at":"2026-05-29T01:05:11Z"},{"alias_kind":"arxiv_version","alias_value":"2605.02772v2","created_at":"2026-05-29T01:05:11Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.02772","created_at":"2026-05-29T01:05:11Z"},{"alias_kind":"pith_short_12","alias_value":"4TT7PDU2ST3K","created_at":"2026-05-29T01:05:11Z"},{"alias_kind":"pith_short_16","alias_value":"4TT7PDU2ST3K2TTC","created_at":"2026-05-29T01:05:11Z"},{"alias_kind":"pith_short_8","alias_value":"4TT7PDU2","created_at":"2026-05-29T01:05:11Z"}],"graph_snapshots":[{"event_id":"sha256:82d00275a1ad24415c3cd7c8a94336aa41a6c7c6c54c3e2e2f8d73d928cf1b67","target":"graph","created_at":"2026-05-29T01:05:11Z","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":"With only 1 hour of fine-tuning on 4×H20 GPUs, SD3.5-T^5 achieves comparable text-to-image quality to the fine-tuned Softmax model, while accelerating inference by 1.32× and 1.47× at 1K and 2K resolutions."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The representational gap between Softmax and linear attention can be closed sufficiently by TTT's two-layer dynamic formulation plus the introduced key instance normalization and lightweight locality enhancement module to allow effective weight inheritance."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Using Test-Time Training's structural match to Softmax attention plus key normalization and locality modules allows inheriting pretrained weights and fine-tuning Stable Diffusion 3.5 in one hour to match quality while speeding inference 1.32-1.47x."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Test-Time Training aligns linear attention with pretrained Softmax weights, enabling transfer after minimal fine-tuning."}],"snapshot_sha256":"93fc46772cf804700f368f4a611d938b6b078ac0e64a8364c7a315c6a6592174"},"formal_canon":{"evidence_count":3,"snapshot_sha256":"dc11f4d94d5680334403414253b47a6a791b65358a24294f407ef69b693e680a"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-20T15:34:05.169126Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_title_agreement","ran_at":"2026-05-20T02:31:22.311632Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-19T15:59:30.513388Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2605.02772/integrity.json","findings":[],"snapshot_sha256":"8a8ce11c3c63c50da39a97018aa53836631ea6d397cc20d8f1ba69798d81f2bb","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"While linear-complexity attention mechanisms offer a promising alternative to Softmax attention for overcoming the quadratic bottleneck, training such models from scratch remains prohibitively expensive. Inheriting weights from pretrained Transformers provides an appealing shortcut, yet the fundamental representational gap between Softmax and linear attention prevents effective weight transfer. In this work, we address this conversion challenge from two perspectives: architectural alignment and representational alignment. We identify Test-Time Training (TTT) as a linear-complexity architecture","authors_text":"Dongchen Han, Gao Huang, Hanyi Wang, Yining Li, Yulin Wang, Zeyu Liu","cross_cats":[],"headline":"Test-Time Training aligns linear attention with pretrained Softmax weights, enabling transfer after minimal fine-tuning.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-05-04T16:16:26Z","title":"Linearizing Vision Transformer with Test-Time Training"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.02772","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-08T18:22:17.491773Z","id":"87624763-74ea-45ed-9738-3fa1bb313768","model_set":{"reader":"grok-4.3"},"one_line_summary":"Using Test-Time Training's structural match to Softmax attention plus key normalization and locality modules allows inheriting pretrained weights and fine-tuning Stable Diffusion 3.5 in one hour to match quality while speeding inference 1.32-1.47x.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Test-Time Training aligns linear attention with pretrained Softmax weights, enabling transfer after minimal fine-tuning.","strongest_claim":"With only 1 hour of fine-tuning on 4×H20 GPUs, SD3.5-T^5 achieves comparable text-to-image quality to the fine-tuned Softmax model, while accelerating inference by 1.32× and 1.47× at 1K and 2K resolutions.","weakest_assumption":"The representational gap between Softmax and linear attention can be closed sufficiently by TTT's two-layer dynamic formulation plus the introduced key instance normalization and lightweight locality enhancement module to allow effective weight inheritance."}},"verdict_id":"87624763-74ea-45ed-9738-3fa1bb313768"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:f8cec3a1fcd99f93aa97b1178ba63e5ba78776beb0b95e4f4f1f6875f996b525","target":"record","created_at":"2026-05-29T01:05:11Z","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":"c70e9ae07c03a95926855e51fc82a37964e7617dc3da0f0f41fa056adff106f7","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2026-05-04T16:16:26Z","title_canon_sha256":"b71a814c742e400a2f004d83bceca10675a1a765736b26042028c90b8cdbaa39"},"schema_version":"1.0","source":{"id":"2605.02772","kind":"arxiv","version":2}},"canonical_sha256":"e4e7f78e9a94f6ad4e6240f43b9f73e9e0fba53aaa694895847f3ea0ef93094b","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"e4e7f78e9a94f6ad4e6240f43b9f73e9e0fba53aaa694895847f3ea0ef93094b","first_computed_at":"2026-05-29T01:05:11.529602Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-29T01:05:11.529602Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"4ZhQi+/R1n0PY69wZE9+ogDCcmxic10bwjjg0D1W8+aWR4tlM+xzbc8DVP08KGPvC+/RBdQ+uGmJ1dVERUsEAQ==","signature_status":"signed_v1","signed_at":"2026-05-29T01:05:11.530905Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.02772","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:f8cec3a1fcd99f93aa97b1178ba63e5ba78776beb0b95e4f4f1f6875f996b525","sha256:82d00275a1ad24415c3cd7c8a94336aa41a6c7c6c54c3e2e2f8d73d928cf1b67"],"state_sha256":"116bf3f2d2bd03fa09e334aa39aaafa2a392e298ba35db3b1e9a2023789c4bf9"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"im+qQG17rVRAhlow3qq/vaS3n4jH0vn0cCQkEDtQyZRYBxImvRRJf7D8ucRpvvQiR6lSPCW2Rj9XvWzyZgdGAw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-06-08T10:40:02.776512Z","bundle_sha256":"b6d0c01757519b9938ba65d2e7ee327eef21d9f694ec23fd596e7e4c156ff6df"}}