{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:WF36QSC6IERIBX3AR7GR2HZCCC","short_pith_number":"pith:WF36QSC6","schema_version":"1.0","canonical_sha256":"b177e8485e412280df608fcd1d1f2210baa80b042bdcfd7133d50112fcd62af8","source":{"kind":"arxiv","id":"2305.14788","version":2},"attestation_state":"computed","paper":{"title":"Adapting Language Models to Compress Contexts","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Alexander Wettig, Alexis Chevalier, Anirudh Ajith, Danqi Chen","submitted_at":"2023-05-24T06:42:44Z","abstract_excerpt":"Transformer-based language models (LMs) are powerful and widely-applicable tools, but their usefulness is constrained by a finite context window and the expensive computational cost of processing long text documents. We propose to adapt pre-trained LMs into AutoCompressors. These language models are capable of compressing long contexts into compact summary vectors, which are then accessible to the model as soft prompts. Summary vectors are trained with an unsupervised objective, whereby long documents are processed in segments, and summary vectors from all previous segments are used in languag"},"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":"2305.14788","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2023-05-24T06:42:44Z","cross_cats_sorted":[],"title_canon_sha256":"22b694fa7ce8c3bc1debc919e7481e89e09c7eb127cabcf5cf61878271e11879","abstract_canon_sha256":"0bc3ac8de9aa045a08620d168b903c334b348b8f788dc5436b2cd13d8473ace2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:08:57.984077Z","signature_b64":"5s2Q07+aY1yK/DpdzBLvqtlQ8VWUuYd5EZrEAUkCZzSU0O8BJ4RP9HRbOck/A1yJjYUt6hnf894d+wJT1mj8DQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b177e8485e412280df608fcd1d1f2210baa80b042bdcfd7133d50112fcd62af8","last_reissued_at":"2026-07-05T07:08:57.983602Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:08:57.983602Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Adapting Language Models to Compress Contexts","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Alexander Wettig, Alexis Chevalier, Anirudh Ajith, Danqi Chen","submitted_at":"2023-05-24T06:42:44Z","abstract_excerpt":"Transformer-based language models (LMs) are powerful and widely-applicable tools, but their usefulness is constrained by a finite context window and the expensive computational cost of processing long text documents. We propose to adapt pre-trained LMs into AutoCompressors. These language models are capable of compressing long contexts into compact summary vectors, which are then accessible to the model as soft prompts. Summary vectors are trained with an unsupervised objective, whereby long documents are processed in segments, and summary vectors from all previous segments are used in languag"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2305.14788","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/2305.14788/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":"2305.14788","created_at":"2026-07-05T07:08:57.983660+00:00"},{"alias_kind":"arxiv_version","alias_value":"2305.14788v2","created_at":"2026-07-05T07:08:57.983660+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2305.14788","created_at":"2026-07-05T07:08:57.983660+00:00"},{"alias_kind":"pith_short_12","alias_value":"WF36QSC6IERI","created_at":"2026-07-05T07:08:57.983660+00:00"},{"alias_kind":"pith_short_16","alias_value":"WF36QSC6IERIBX3A","created_at":"2026-07-05T07:08:57.983660+00:00"},{"alias_kind":"pith_short_8","alias_value":"WF36QSC6","created_at":"2026-07-05T07:08:57.983660+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":17,"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":20,"is_internal_anchor":true},{"citing_arxiv_id":"2606.20295","citing_title":"Token-Operations-Oriented Inference Optimization Techniques for Large Models","ref_index":127,"is_internal_anchor":false},{"citing_arxiv_id":"2606.11853","citing_title":"Task-Aware Structured Memory for Dynamic Multi-modal In-Context Learning","ref_index":209,"is_internal_anchor":false},{"citing_arxiv_id":"2606.09659","citing_title":"End-to-End Context Compression at Scale","ref_index":11,"is_internal_anchor":false},{"citing_arxiv_id":"2606.03979","citing_title":"Language Models Need Sleep: Learning to Self-Modify and Consolidate Memories","ref_index":53,"is_internal_anchor":false},{"citing_arxiv_id":"2606.20571","citing_title":"Less is More: Lightweight Prompt Compression for Question Answering Applications on Edge Devices","ref_index":76,"is_internal_anchor":false},{"citing_arxiv_id":"2605.22337","citing_title":"Meta-Soft: Leveraging Composable Meta-Tokens for Context-Preserving KV Cache Compression","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2409.01579","citing_title":"AdaComp: Extractive Context Compression with Adaptive Predictor for Retrieval-Augmented Large Language Models","ref_index":4,"is_internal_anchor":false},{"citing_arxiv_id":"2501.19201","citing_title":"Efficient Reasoning with Hidden Thinking","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2502.14925","citing_title":"CODEPROMPTZIP: Code-specific Prompt Compression for Retrieval-Augmented Generation in Coding Tasks with LMs","ref_index":3,"is_internal_anchor":false},{"citing_arxiv_id":"2605.22337","citing_title":"Meta-Soft: Leveraging Composable Meta-Tokens for Context-Preserving KV Cache Compression","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2404.07143","citing_title":"Leave No Context Behind: Efficient Infinite Context Transformers with Infini-attention","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2508.16745","citing_title":"Beyond Memorization: Extending Reasoning Depth with Recurrence, Memory and Test-Time Compute Scaling","ref_index":10,"is_internal_anchor":false},{"citing_arxiv_id":"2401.05459","citing_title":"Personal LLM Agents: Insights and Survey about the Capability, Efficiency and Security","ref_index":264,"is_internal_anchor":false},{"citing_arxiv_id":"2604.09607","citing_title":"LLM-assisted Agentic Edge Intelligence Framework","ref_index":20,"is_internal_anchor":false},{"citing_arxiv_id":"2404.14294","citing_title":"A Survey on Efficient Inference for Large Language Models","ref_index":30,"is_internal_anchor":false},{"citing_arxiv_id":"2604.26483","citing_title":"Efficient Listwise Reranking with Compressed Document Representations","ref_index":6,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/WF36QSC6IERIBX3AR7GR2HZCCC","json":"https://pith.science/pith/WF36QSC6IERIBX3AR7GR2HZCCC.json","graph_json":"https://pith.science/api/pith-number/WF36QSC6IERIBX3AR7GR2HZCCC/graph.json","events_json":"https://pith.science/api/pith-number/WF36QSC6IERIBX3AR7GR2HZCCC/events.json","paper":"https://pith.science/paper/WF36QSC6"},"agent_actions":{"view_html":"https://pith.science/pith/WF36QSC6IERIBX3AR7GR2HZCCC","download_json":"https://pith.science/pith/WF36QSC6IERIBX3AR7GR2HZCCC.json","view_paper":"https://pith.science/paper/WF36QSC6","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2305.14788&json=true","fetch_graph":"https://pith.science/api/pith-number/WF36QSC6IERIBX3AR7GR2HZCCC/graph.json","fetch_events":"https://pith.science/api/pith-number/WF36QSC6IERIBX3AR7GR2HZCCC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WF36QSC6IERIBX3AR7GR2HZCCC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WF36QSC6IERIBX3AR7GR2HZCCC/action/storage_attestation","attest_author":"https://pith.science/pith/WF36QSC6IERIBX3AR7GR2HZCCC/action/author_attestation","sign_citation":"https://pith.science/pith/WF36QSC6IERIBX3AR7GR2HZCCC/action/citation_signature","submit_replication":"https://pith.science/pith/WF36QSC6IERIBX3AR7GR2HZCCC/action/replication_record"}},"created_at":"2026-07-05T07:08:57.983660+00:00","updated_at":"2026-07-05T07:08:57.983660+00:00"}