{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:ZUOGOBDSLEEGKLN7KE3D43VL5H","short_pith_number":"pith:ZUOGOBDS","schema_version":"1.0","canonical_sha256":"cd1c6704725908652dbf51363e6eabe9fbd579274d1cd93a5ff437215156c803","source":{"kind":"arxiv","id":"2607.06827","version":1},"attestation_state":"computed","paper":{"title":"Compress the Cache, Not the Speech Embedding: KV Compression for Efficient Speech LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.SD"],"primary_cat":"eess.AS","authors_text":"Jinyu Li, Ke-Han Lu, Keqi Deng, Ruchao Fan, Rui Zhao","submitted_at":"2026-07-07T21:45:04Z","abstract_excerpt":"Speech large language models (Speech LLMs) typically encode speech into sequences far longer than text, creating a major efficiency bottleneck during autoregressive decoding. A common remedy is to compress the speech sequence at the adapter level to remove temporal redundancy before it enters the LLM; however, such early downsampling risks discarding fine-grained information that cannot be recovered. We propose SpeechKV, which applies a learned pooling to the KV cache of speech tokens inside the LLM. This design allows the LLM to fuse speech and text internally while directly accelerating deco"},"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":"2607.06827","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"eess.AS","submitted_at":"2026-07-07T21:45:04Z","cross_cats_sorted":["cs.SD"],"title_canon_sha256":"245f4c687e8e10aba3ac1bd6ff19b01ad8f7caba534bc39f0db6ad7747d2daf3","abstract_canon_sha256":"61506fa8372a47c5b2685760c608ff81dd3ece4382b3f679da8c5e79a9e753ee"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-09T00:19:29.883221Z","signature_b64":"/FiX1ax2cC93ovZTG//7yUM6lWNA2wIS+hidZ+jauPvMuAN4gO/UtH0bfSLVciyViBCU5eTbnpzwu+Wy3rgyCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"cd1c6704725908652dbf51363e6eabe9fbd579274d1cd93a5ff437215156c803","last_reissued_at":"2026-07-09T00:19:29.882424Z","signature_status":"signed_v1","first_computed_at":"2026-07-09T00:19:29.882424Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Compress the Cache, Not the Speech Embedding: KV Compression for Efficient Speech LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.SD"],"primary_cat":"eess.AS","authors_text":"Jinyu Li, Ke-Han Lu, Keqi Deng, Ruchao Fan, Rui Zhao","submitted_at":"2026-07-07T21:45:04Z","abstract_excerpt":"Speech large language models (Speech LLMs) typically encode speech into sequences far longer than text, creating a major efficiency bottleneck during autoregressive decoding. A common remedy is to compress the speech sequence at the adapter level to remove temporal redundancy before it enters the LLM; however, such early downsampling risks discarding fine-grained information that cannot be recovered. We propose SpeechKV, which applies a learned pooling to the KV cache of speech tokens inside the LLM. This design allows the LLM to fuse speech and text internally while directly accelerating deco"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.06827","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/2607.06827/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":"2607.06827","created_at":"2026-07-09T00:19:29.882584+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.06827v1","created_at":"2026-07-09T00:19:29.882584+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.06827","created_at":"2026-07-09T00:19:29.882584+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZUOGOBDSLEEG","created_at":"2026-07-09T00:19:29.882584+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZUOGOBDSLEEGKLN7","created_at":"2026-07-09T00:19:29.882584+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZUOGOBDS","created_at":"2026-07-09T00:19:29.882584+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ZUOGOBDSLEEGKLN7KE3D43VL5H","json":"https://pith.science/pith/ZUOGOBDSLEEGKLN7KE3D43VL5H.json","graph_json":"https://pith.science/api/pith-number/ZUOGOBDSLEEGKLN7KE3D43VL5H/graph.json","events_json":"https://pith.science/api/pith-number/ZUOGOBDSLEEGKLN7KE3D43VL5H/events.json","paper":"https://pith.science/paper/ZUOGOBDS"},"agent_actions":{"view_html":"https://pith.science/pith/ZUOGOBDSLEEGKLN7KE3D43VL5H","download_json":"https://pith.science/pith/ZUOGOBDSLEEGKLN7KE3D43VL5H.json","view_paper":"https://pith.science/paper/ZUOGOBDS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.06827&json=true","fetch_graph":"https://pith.science/api/pith-number/ZUOGOBDSLEEGKLN7KE3D43VL5H/graph.json","fetch_events":"https://pith.science/api/pith-number/ZUOGOBDSLEEGKLN7KE3D43VL5H/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZUOGOBDSLEEGKLN7KE3D43VL5H/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZUOGOBDSLEEGKLN7KE3D43VL5H/action/storage_attestation","attest_author":"https://pith.science/pith/ZUOGOBDSLEEGKLN7KE3D43VL5H/action/author_attestation","sign_citation":"https://pith.science/pith/ZUOGOBDSLEEGKLN7KE3D43VL5H/action/citation_signature","submit_replication":"https://pith.science/pith/ZUOGOBDSLEEGKLN7KE3D43VL5H/action/replication_record"}},"created_at":"2026-07-09T00:19:29.882584+00:00","updated_at":"2026-07-09T00:19:29.882584+00:00"}