{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:PZVMPK2RKJCXXVJHEQNMFXE2LW","short_pith_number":"pith:PZVMPK2R","schema_version":"1.0","canonical_sha256":"7e6ac7ab5152457bd527241ac2dc9a5dbb9bac366d35dd33e2ec19c077c5d93d","source":{"kind":"arxiv","id":"2401.06104","version":2},"attestation_state":"computed","paper":{"title":"Transformers are Multi-State RNNs","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Matanel Oren, Michael Hassid, Nir Yarden, Roy Schwartz, Yossi Adi","submitted_at":"2024-01-11T18:35:26Z","abstract_excerpt":"Transformers are considered conceptually different from the previous generation of state-of-the-art NLP models - recurrent neural networks (RNNs). In this work, we demonstrate that decoder-only transformers can in fact be conceptualized as unbounded multi-state RNNs - an RNN variant with unlimited hidden state size. We further show that transformers can be converted into $\\textit{bounded}$ multi-state RNNs by fixing the size of their hidden state, effectively compressing their key-value cache. We introduce a novel, training-free compression policy - $\\textbf{T}$oken $\\textbf{O}$mission $\\textb"},"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":"2401.06104","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2024-01-11T18:35:26Z","cross_cats_sorted":[],"title_canon_sha256":"9d6ddac89273db38ad440e5e6790f3b676e61584224db0d80bc9704e0f78eaa2","abstract_canon_sha256":"8d41cf8cb05fadb5b03b6e2e9c627ba541d906378b755a4dcf67ba8e399c212e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:33:34.408289Z","signature_b64":"0J1NKcsm+ofoZs0z2WlgLNgN+JzmsLZvPY1olZ4sDCTNTOhqu6FEmo5k1ezpaM8dKitqx316r0lh+48O1FwkAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7e6ac7ab5152457bd527241ac2dc9a5dbb9bac366d35dd33e2ec19c077c5d93d","last_reissued_at":"2026-07-05T08:33:34.407678Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:33:34.407678Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Transformers are Multi-State RNNs","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Matanel Oren, Michael Hassid, Nir Yarden, Roy Schwartz, Yossi Adi","submitted_at":"2024-01-11T18:35:26Z","abstract_excerpt":"Transformers are considered conceptually different from the previous generation of state-of-the-art NLP models - recurrent neural networks (RNNs). In this work, we demonstrate that decoder-only transformers can in fact be conceptualized as unbounded multi-state RNNs - an RNN variant with unlimited hidden state size. We further show that transformers can be converted into $\\textit{bounded}$ multi-state RNNs by fixing the size of their hidden state, effectively compressing their key-value cache. We introduce a novel, training-free compression policy - $\\textbf{T}$oken $\\textbf{O}$mission $\\textb"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2401.06104","kind":"arxiv","version":2},"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/2401.06104/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":"2401.06104","created_at":"2026-07-05T08:33:34.407751+00:00"},{"alias_kind":"arxiv_version","alias_value":"2401.06104v2","created_at":"2026-07-05T08:33:34.407751+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2401.06104","created_at":"2026-07-05T08:33:34.407751+00:00"},{"alias_kind":"pith_short_12","alias_value":"PZVMPK2RKJCX","created_at":"2026-07-05T08:33:34.407751+00:00"},{"alias_kind":"pith_short_16","alias_value":"PZVMPK2RKJCXXVJH","created_at":"2026-07-05T08:33:34.407751+00:00"},{"alias_kind":"pith_short_8","alias_value":"PZVMPK2R","created_at":"2026-07-05T08:33:34.407751+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":15,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.08032","citing_title":"What to Keep, What to Forget: A Rate--Distortion View of Memory Compaction in LLMs and Agents","ref_index":89,"is_internal_anchor":true},{"citing_arxiv_id":"2606.24957","citing_title":"Dustin: Draft-Augmented Sparse Verification for Efficient Long-Context Generation with Speculative Decoding","ref_index":7,"is_internal_anchor":false},{"citing_arxiv_id":"2606.08302","citing_title":"HACK++: Towards More Effective Head-Aware Key-Value Compression for Efficient Visual Autoregressive Modeling","ref_index":43,"is_internal_anchor":false},{"citing_arxiv_id":"2605.25475","citing_title":"IndexMem: Learned KV-Cache Eviction with Latent Memory for Long-Context LLM Inference","ref_index":19,"is_internal_anchor":false},{"citing_arxiv_id":"2602.18196","citing_title":"RAT+: Train Dense, Infer Sparse -- Recurrence Augmented Attention for Dilated Inference","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2510.00231","citing_title":"The Pitfalls of KV Cache Compression","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2510.09883","citing_title":"DELTA: Dynamic Layer-Aware Token Attention for Efficient Long-Context Reasoning","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2502.13189","citing_title":"MoBA: Mixture of Block Attention for Long-Context LLMs","ref_index":28,"is_internal_anchor":false},{"citing_arxiv_id":"2602.18196","citing_title":"RAT+: Train Dense, Infer Sparse -- Recurrence Augmented Attention for Dilated Inference","ref_index":24,"is_internal_anchor":false},{"citing_arxiv_id":"2406.10774","citing_title":"Quest: Query-Aware Sparsity for Efficient Long-Context LLM Inference","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2312.06635","citing_title":"Gated Linear Attention Transformers with Hardware-Efficient Training","ref_index":61,"is_internal_anchor":false},{"citing_arxiv_id":"2605.08840","citing_title":"ReST-KV: Robust KV Cache Eviction with Layer-wise Output Reconstruction and Spatial-Temporal Smoothing","ref_index":17,"is_internal_anchor":false},{"citing_arxiv_id":"2605.06554","citing_title":"Long Context Pre-Training with Lighthouse Attention","ref_index":26,"is_internal_anchor":false},{"citing_arxiv_id":"2604.18137","citing_title":"AQPIM: Breaking the PIM Capacity Wall for LLMs with In-Memory Activation Quantization","ref_index":57,"is_internal_anchor":false},{"citing_arxiv_id":"2604.11288","citing_title":"Transactional Attention: Semantic Sponsorship for KV-Cache Retention","ref_index":18,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/PZVMPK2RKJCXXVJHEQNMFXE2LW","json":"https://pith.science/pith/PZVMPK2RKJCXXVJHEQNMFXE2LW.json","graph_json":"https://pith.science/api/pith-number/PZVMPK2RKJCXXVJHEQNMFXE2LW/graph.json","events_json":"https://pith.science/api/pith-number/PZVMPK2RKJCXXVJHEQNMFXE2LW/events.json","paper":"https://pith.science/paper/PZVMPK2R"},"agent_actions":{"view_html":"https://pith.science/pith/PZVMPK2RKJCXXVJHEQNMFXE2LW","download_json":"https://pith.science/pith/PZVMPK2RKJCXXVJHEQNMFXE2LW.json","view_paper":"https://pith.science/paper/PZVMPK2R","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2401.06104&json=true","fetch_graph":"https://pith.science/api/pith-number/PZVMPK2RKJCXXVJHEQNMFXE2LW/graph.json","fetch_events":"https://pith.science/api/pith-number/PZVMPK2RKJCXXVJHEQNMFXE2LW/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PZVMPK2RKJCXXVJHEQNMFXE2LW/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PZVMPK2RKJCXXVJHEQNMFXE2LW/action/storage_attestation","attest_author":"https://pith.science/pith/PZVMPK2RKJCXXVJHEQNMFXE2LW/action/author_attestation","sign_citation":"https://pith.science/pith/PZVMPK2RKJCXXVJHEQNMFXE2LW/action/citation_signature","submit_replication":"https://pith.science/pith/PZVMPK2RKJCXXVJHEQNMFXE2LW/action/replication_record"}},"created_at":"2026-07-05T08:33:34.407751+00:00","updated_at":"2026-07-05T08:33:34.407751+00:00"}