{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:VNFUHXCLJAPFGIGBHHUNERJ47Y","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":"1ef38dabdc5d4752e11d311d9445b475b24b4c6ac345155aa2f530c37a22414c","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-14T14:22:34Z","title_canon_sha256":"4b56ae184d7afff6d28f203935951a47bae3d761028d3320a9f5096242185147"},"schema_version":"1.0","source":{"id":"2605.14877","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.14877","created_at":"2026-05-17T23:38:56Z"},{"alias_kind":"arxiv_version","alias_value":"2605.14877v1","created_at":"2026-05-17T23:38:56Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.14877","created_at":"2026-05-17T23:38:56Z"},{"alias_kind":"pith_short_12","alias_value":"VNFUHXCLJAPF","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"VNFUHXCLJAPFGIGB","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"VNFUHXCL","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:b6a8e0672680af75fd4b4832c10954b99479a659316176a00a2fc8c92f55ede6","target":"graph","created_at":"2026-05-17T23:38:56Z","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":"Applied to the Infinity-2B model, HeatKV achieves 2× higher compression ratio in memory allocation for KV cache compared to existing methods, while maintaining similar or better image fidelity, prompt alignment and human perception score."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That a static pruning schedule derived from attention scores on a small offline calibration set will generalize to arbitrary prompts and generation lengths without measurable quality loss."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"HeatKV ranks attention heads by their focus on prior scales using offline calibration data and applies a static per-head pruning schedule, delivering 2x higher KV-cache compression than prior methods on the Infinity-2B model with comparable image fidelity."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Head-specific attention ranking doubles KV-cache compression in visual autoregressive image models while preserving quality."}],"snapshot_sha256":"65ba968a434d962722aa0603e36d76022c7c8a7be8960fab444d48979d528093"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"63aeb8692867d4ce5341a164139d63f584be7c504aea060e9cd9ca6f67647c72"},"paper":{"abstract_excerpt":"Visual Autoregressive (VAR) models have recently demonstrated impressive image generation quality while maintaining low latency. However, they suffer from severe KV-cache memory constraints, often requiring gigabytes of memory per generated image. We introduce HeatKV, a novel compression method that adapts cache allocation in each head based on its attention to previously generated scales. Using a small offline calibration set, the attention heads are ranked according to their attention scores over prior scales. Based on this ranking, we construct a static pruning schedule tailored to a given ","authors_text":"Axel Berg, Chuteng Zhou, Durmus Alp Emre Acar, Jonathan Cederlund, Pontus Giselsson","cross_cats":[],"headline":"Head-specific attention ranking doubles KV-cache compression in visual autoregressive image models while preserving quality.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-14T14:22:34Z","title":"HeatKV: Head-tuned KV-cache Compression for Visual Autoregressive Modeling"},"references":{"count":40,"internal_anchors":2,"resolved_work":40,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Infinity: Scaling bitwise autoregressive modeling for high-resolution image synthesis,","work_id":"977492f9-1add-4df5-aa71-c2e7f1a694b6","year":2025},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"I. J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y . Bengio, “Generative adversarial nets,”Advances in neural information processing systems, vol. 27, ","work_id":"4b8320ae-4a24-4686-85ca-93e608f3af08","year":2014},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"A style-based generator architecture for generative adversarial networks,","work_id":"db3bde24-03b8-4609-98c5-e9672550b362","year":2019},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Stylegan-xl: Scaling stylegan to large diverse datasets,","work_id":"f5a2bd2d-af70-447d-a377-f8717494b10f","year":2022},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Diffusion models beat gans on image synthesis,","work_id":"35493ec7-8582-4e9b-a3dc-d835aca70a28","year":2021}],"snapshot_sha256":"bb5685a19cb18a9b4eee104188b34d8e9606abdd26d8b09aefd67fe8dfe4d354"},"source":{"id":"2605.14877","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-15T03:15:13.332676Z","id":"fb93b85d-6058-4875-89dc-c22b53b3a353","model_set":{"reader":"grok-4.3"},"one_line_summary":"HeatKV ranks attention heads by their focus on prior scales using offline calibration data and applies a static per-head pruning schedule, delivering 2x higher KV-cache compression than prior methods on the Infinity-2B model with comparable image fidelity.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Head-specific attention ranking doubles KV-cache compression in visual autoregressive image models while preserving quality.","strongest_claim":"Applied to the Infinity-2B model, HeatKV achieves 2× higher compression ratio in memory allocation for KV cache compared to existing methods, while maintaining similar or better image fidelity, prompt alignment and human perception score.","weakest_assumption":"That a static pruning schedule derived from attention scores on a small offline calibration set will generalize to arbitrary prompts and generation lengths without measurable quality loss."}},"verdict_id":"fb93b85d-6058-4875-89dc-c22b53b3a353"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:8d4dd7576ee3e9c71b7edf89b061b9dfb4122239f74e9f1e0148c78c4b86e4be","target":"record","created_at":"2026-05-17T23:38:56Z","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":"1ef38dabdc5d4752e11d311d9445b475b24b4c6ac345155aa2f530c37a22414c","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-05-14T14:22:34Z","title_canon_sha256":"4b56ae184d7afff6d28f203935951a47bae3d761028d3320a9f5096242185147"},"schema_version":"1.0","source":{"id":"2605.14877","kind":"arxiv","version":1}},"canonical_sha256":"ab4b43dc4b481e5320c139e8d2453cfe073be42803dbf4c2e4b9c684cad8d1ea","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"ab4b43dc4b481e5320c139e8d2453cfe073be42803dbf4c2e4b9c684cad8d1ea","first_computed_at":"2026-05-17T23:38:56.081842Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:38:56.081842Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"TjdIM4m7hPbQc41BQbHDJSuvBKztPZI0N3Kd9c+JjqNkQIaySmoSj/pzlYnmKWjIvsbIj2D4a52TLGdbIZq0DA==","signature_status":"signed_v1","signed_at":"2026-05-17T23:38:56.082506Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.14877","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:8d4dd7576ee3e9c71b7edf89b061b9dfb4122239f74e9f1e0148c78c4b86e4be","sha256:b6a8e0672680af75fd4b4832c10954b99479a659316176a00a2fc8c92f55ede6"],"state_sha256":"3c4ded680787a2ffd7263143e8c480166aae25e9855ce932837c19a5c8e1e1dd"}