{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:PXBAZMLXDUJVQED25E7GFRDS3V","short_pith_number":"pith:PXBAZMLX","schema_version":"1.0","canonical_sha256":"7dc20cb1771d1358107ae93e62c472dd5189b6c2c09e9fd12546557126bedb9d","source":{"kind":"arxiv","id":"2503.23274","version":1},"attestation_state":"computed","paper":{"title":"PromptDistill: Query-based Selective Token Retention in Intermediate Layers for Efficient Large Language Model Inference","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Amir Zadeh, Chuan Li, Maojia Song, Soujanya Poria, Tej Deep Pala, Weisheng Jin, Yew Ken Chia","submitted_at":"2025-03-30T01:47:23Z","abstract_excerpt":"As large language models (LLMs) tackle increasingly complex tasks and longer documents, their computational and memory costs during inference become a major bottleneck. To address this, we propose PromptDistill, a novel, training-free method that improves inference efficiency while preserving generation quality. PromptDistill identifies and retains the most informative tokens by leveraging attention interactions in early layers, preserving their hidden states while reducing the computational burden in later layers. This allows the model to focus on essential contextual information without full"},"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":"2503.23274","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2025-03-30T01:47:23Z","cross_cats_sorted":[],"title_canon_sha256":"183debc2c0b340aa93fcca121c17b4ae8d8fcd8e0cc4e9c21a548de4baa19874","abstract_canon_sha256":"1e6085840233c731278325e2a5fff3f62125e5d71d17d32df3333eb0923c5d41"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:41:33.282129Z","signature_b64":"ZK/WkScPm8q2gzU387zwYTXaNwypRd2AKJszWR5+1DJ+C11/rXpE5AnF3VmnGCaEFp5+a3ROsR/CQq0OyX5fCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"7dc20cb1771d1358107ae93e62c472dd5189b6c2c09e9fd12546557126bedb9d","last_reissued_at":"2026-07-05T10:41:33.281641Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:41:33.281641Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"PromptDistill: Query-based Selective Token Retention in Intermediate Layers for Efficient Large Language Model Inference","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Amir Zadeh, Chuan Li, Maojia Song, Soujanya Poria, Tej Deep Pala, Weisheng Jin, Yew Ken Chia","submitted_at":"2025-03-30T01:47:23Z","abstract_excerpt":"As large language models (LLMs) tackle increasingly complex tasks and longer documents, their computational and memory costs during inference become a major bottleneck. To address this, we propose PromptDistill, a novel, training-free method that improves inference efficiency while preserving generation quality. PromptDistill identifies and retains the most informative tokens by leveraging attention interactions in early layers, preserving their hidden states while reducing the computational burden in later layers. This allows the model to focus on essential contextual information without full"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.23274","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/2503.23274/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":"2503.23274","created_at":"2026-07-05T10:41:33.281709+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.23274v1","created_at":"2026-07-05T10:41:33.281709+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.23274","created_at":"2026-07-05T10:41:33.281709+00:00"},{"alias_kind":"pith_short_12","alias_value":"PXBAZMLXDUJV","created_at":"2026-07-05T10:41:33.281709+00:00"},{"alias_kind":"pith_short_16","alias_value":"PXBAZMLXDUJVQED2","created_at":"2026-07-05T10:41:33.281709+00:00"},{"alias_kind":"pith_short_8","alias_value":"PXBAZMLX","created_at":"2026-07-05T10:41:33.281709+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/PXBAZMLXDUJVQED25E7GFRDS3V","json":"https://pith.science/pith/PXBAZMLXDUJVQED25E7GFRDS3V.json","graph_json":"https://pith.science/api/pith-number/PXBAZMLXDUJVQED25E7GFRDS3V/graph.json","events_json":"https://pith.science/api/pith-number/PXBAZMLXDUJVQED25E7GFRDS3V/events.json","paper":"https://pith.science/paper/PXBAZMLX"},"agent_actions":{"view_html":"https://pith.science/pith/PXBAZMLXDUJVQED25E7GFRDS3V","download_json":"https://pith.science/pith/PXBAZMLXDUJVQED25E7GFRDS3V.json","view_paper":"https://pith.science/paper/PXBAZMLX","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.23274&json=true","fetch_graph":"https://pith.science/api/pith-number/PXBAZMLXDUJVQED25E7GFRDS3V/graph.json","fetch_events":"https://pith.science/api/pith-number/PXBAZMLXDUJVQED25E7GFRDS3V/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/PXBAZMLXDUJVQED25E7GFRDS3V/action/timestamp_anchor","attest_storage":"https://pith.science/pith/PXBAZMLXDUJVQED25E7GFRDS3V/action/storage_attestation","attest_author":"https://pith.science/pith/PXBAZMLXDUJVQED25E7GFRDS3V/action/author_attestation","sign_citation":"https://pith.science/pith/PXBAZMLXDUJVQED25E7GFRDS3V/action/citation_signature","submit_replication":"https://pith.science/pith/PXBAZMLXDUJVQED25E7GFRDS3V/action/replication_record"}},"created_at":"2026-07-05T10:41:33.281709+00:00","updated_at":"2026-07-05T10:41:33.281709+00:00"}