{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:HM5I6NU55YO2XXCP6I74VQQH7E","short_pith_number":"pith:HM5I6NU5","schema_version":"1.0","canonical_sha256":"3b3a8f369dee1dabdc4ff23fcac207f9356c81e834a1f2152f5edde2b50cf7f1","source":{"kind":"arxiv","id":"2412.17747","version":1},"attestation_state":"computed","paper":{"title":"Deliberation in Latent Space via Differentiable Cache Augmentation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Arthur Szlam, Jiaxing Wu, Jonas Pfeiffer, Jun Xie, Luyang Liu","submitted_at":"2024-12-23T18:02:25Z","abstract_excerpt":"Techniques enabling large language models (LLMs) to \"think more\" by generating and attending to intermediate reasoning steps have shown promise in solving complex problems. However, the standard approaches generate sequences of discrete tokens immediately before responding, and so they can incur significant latency costs and be challenging to optimize. In this work, we demonstrate that a frozen LLM can be augmented with an offline coprocessor that operates on the model's key-value (kv) cache. This coprocessor augments the cache with a set of latent embeddings designed to improve the fidelity o"},"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":"2412.17747","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-12-23T18:02:25Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"ee34558fad14225a5de9ca820bfeb323df3bed802095dc7a861b1ffb26ec1f87","abstract_canon_sha256":"018efa56b0385757b2ba0a26b874813b6a19fd7fa50fbc73634eb1b0bcbdb560"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:53:26.874149Z","signature_b64":"ccLPuBe1M6yQNks1STKrWoev9RB0i8oeqoemRmV6hRUtxrBC/pmxisHLjkcF3RAHSfb1mU5zWCVLkMSq7cMyDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"3b3a8f369dee1dabdc4ff23fcac207f9356c81e834a1f2152f5edde2b50cf7f1","last_reissued_at":"2026-07-05T09:53:26.873725Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:53:26.873725Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Deliberation in Latent Space via Differentiable Cache Augmentation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Arthur Szlam, Jiaxing Wu, Jonas Pfeiffer, Jun Xie, Luyang Liu","submitted_at":"2024-12-23T18:02:25Z","abstract_excerpt":"Techniques enabling large language models (LLMs) to \"think more\" by generating and attending to intermediate reasoning steps have shown promise in solving complex problems. However, the standard approaches generate sequences of discrete tokens immediately before responding, and so they can incur significant latency costs and be challenging to optimize. In this work, we demonstrate that a frozen LLM can be augmented with an offline coprocessor that operates on the model's key-value (kv) cache. This coprocessor augments the cache with a set of latent embeddings designed to improve the fidelity o"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2412.17747","kind":"arxiv","version":1},"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/2412.17747/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":"2412.17747","created_at":"2026-07-05T09:53:26.873783+00:00"},{"alias_kind":"arxiv_version","alias_value":"2412.17747v1","created_at":"2026-07-05T09:53:26.873783+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2412.17747","created_at":"2026-07-05T09:53:26.873783+00:00"},{"alias_kind":"pith_short_12","alias_value":"HM5I6NU55YO2","created_at":"2026-07-05T09:53:26.873783+00:00"},{"alias_kind":"pith_short_16","alias_value":"HM5I6NU55YO2XXCP","created_at":"2026-07-05T09:53:26.873783+00:00"},{"alias_kind":"pith_short_8","alias_value":"HM5I6NU5","created_at":"2026-07-05T09:53:26.873783+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2604.26283","citing_title":"MedSynapse-V: Bridging Visual Perception and Clinical Intuition via Latent Memory Evolution","ref_index":73,"is_internal_anchor":false},{"citing_arxiv_id":"2604.26283","citing_title":"MedSynapse-V: Bridging Visual Perception and Clinical Intuition via Latent Memory Evolution","ref_index":73,"is_internal_anchor":false},{"citing_arxiv_id":"2604.26283","citing_title":"MedSynapse-V: Bridging Visual Perception and Clinical Intuition via Latent Memory Evolution","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2502.05171","citing_title":"Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach","ref_index":99,"is_internal_anchor":false},{"citing_arxiv_id":"2604.26283","citing_title":"MedSynapse-V: Bridging Visual Perception and Clinical Intuition via Latent Memory Evolution","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2604.25642","citing_title":"Prefill-Time Intervention for Mitigating Hallucination in Large Vision-Language Models","ref_index":28,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/HM5I6NU55YO2XXCP6I74VQQH7E","json":"https://pith.science/pith/HM5I6NU55YO2XXCP6I74VQQH7E.json","graph_json":"https://pith.science/api/pith-number/HM5I6NU55YO2XXCP6I74VQQH7E/graph.json","events_json":"https://pith.science/api/pith-number/HM5I6NU55YO2XXCP6I74VQQH7E/events.json","paper":"https://pith.science/paper/HM5I6NU5"},"agent_actions":{"view_html":"https://pith.science/pith/HM5I6NU55YO2XXCP6I74VQQH7E","download_json":"https://pith.science/pith/HM5I6NU55YO2XXCP6I74VQQH7E.json","view_paper":"https://pith.science/paper/HM5I6NU5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2412.17747&json=true","fetch_graph":"https://pith.science/api/pith-number/HM5I6NU55YO2XXCP6I74VQQH7E/graph.json","fetch_events":"https://pith.science/api/pith-number/HM5I6NU55YO2XXCP6I74VQQH7E/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/HM5I6NU55YO2XXCP6I74VQQH7E/action/timestamp_anchor","attest_storage":"https://pith.science/pith/HM5I6NU55YO2XXCP6I74VQQH7E/action/storage_attestation","attest_author":"https://pith.science/pith/HM5I6NU55YO2XXCP6I74VQQH7E/action/author_attestation","sign_citation":"https://pith.science/pith/HM5I6NU55YO2XXCP6I74VQQH7E/action/citation_signature","submit_replication":"https://pith.science/pith/HM5I6NU55YO2XXCP6I74VQQH7E/action/replication_record"}},"created_at":"2026-07-05T09:53:26.873783+00:00","updated_at":"2026-07-05T09:53:26.873783+00:00"}