{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:KBNXNJEHRQLBSYHNZT46E3P7WU","short_pith_number":"pith:KBNXNJEH","schema_version":"1.0","canonical_sha256":"505b76a4878c161960edccf9e26dffb5283b805c7f4884737421ba703ebbd3ee","source":{"kind":"arxiv","id":"2205.12674","version":3},"attestation_state":"computed","paper":{"title":"Training Language Models with Memory Augmentation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Danqi Chen, Tao Lei, Zexuan Zhong","submitted_at":"2022-05-25T11:37:29Z","abstract_excerpt":"Recent work has improved language models (LMs) remarkably by equipping them with a non-parametric memory component. However, most existing approaches only introduce mem-ories at testing time or represent them using a separately trained encoder, resulting in suboptimal training of the language model. In this work, we present TRIME, a novel yet simple training approach designed for training LMs with memory augmentation. Our approach uses a training objective that directly takes in-batch examples as accessible memory. We also present new methods for memory construction and data batching, which ar"},"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":"2205.12674","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2022-05-25T11:37:29Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"4ee1083098ebc6f8c4f5f98936acbe862ddebb4dc25b3c841fe7b035a1d28968","abstract_canon_sha256":"d67d4b949506bd7b1944eaed7cd0a16dc74fdaa9c8e46a82d8a34c36f99bedfa"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:20:11.771205Z","signature_b64":"nOIQIxntB/iKfK1OYYWCHrioDSOqpiJr6yQMMVsHOgod1jgDkjZZuFMitYR3ss/izsfCVXAHEUOI4ns3MLi9Dw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"505b76a4878c161960edccf9e26dffb5283b805c7f4884737421ba703ebbd3ee","last_reissued_at":"2026-07-05T05:20:11.770692Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:20:11.770692Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Training Language Models with Memory Augmentation","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Danqi Chen, Tao Lei, Zexuan Zhong","submitted_at":"2022-05-25T11:37:29Z","abstract_excerpt":"Recent work has improved language models (LMs) remarkably by equipping them with a non-parametric memory component. However, most existing approaches only introduce mem-ories at testing time or represent them using a separately trained encoder, resulting in suboptimal training of the language model. In this work, we present TRIME, a novel yet simple training approach designed for training LMs with memory augmentation. Our approach uses a training objective that directly takes in-batch examples as accessible memory. We also present new methods for memory construction and data batching, which ar"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2205.12674","kind":"arxiv","version":3},"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/2205.12674/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":"2205.12674","created_at":"2026-07-05T05:20:11.770754+00:00"},{"alias_kind":"arxiv_version","alias_value":"2205.12674v3","created_at":"2026-07-05T05:20:11.770754+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2205.12674","created_at":"2026-07-05T05:20:11.770754+00:00"},{"alias_kind":"pith_short_12","alias_value":"KBNXNJEHRQLB","created_at":"2026-07-05T05:20:11.770754+00:00"},{"alias_kind":"pith_short_16","alias_value":"KBNXNJEHRQLBSYHN","created_at":"2026-07-05T05:20:11.770754+00:00"},{"alias_kind":"pith_short_8","alias_value":"KBNXNJEH","created_at":"2026-07-05T05:20:11.770754+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.11445","citing_title":"Forecasting Future Behavior as a Learning Task","ref_index":83,"is_internal_anchor":false},{"citing_arxiv_id":"2604.19899","citing_title":"A Reproducibility Study of Metacognitive Retrieval-Augmented Generation","ref_index":58,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/KBNXNJEHRQLBSYHNZT46E3P7WU","json":"https://pith.science/pith/KBNXNJEHRQLBSYHNZT46E3P7WU.json","graph_json":"https://pith.science/api/pith-number/KBNXNJEHRQLBSYHNZT46E3P7WU/graph.json","events_json":"https://pith.science/api/pith-number/KBNXNJEHRQLBSYHNZT46E3P7WU/events.json","paper":"https://pith.science/paper/KBNXNJEH"},"agent_actions":{"view_html":"https://pith.science/pith/KBNXNJEHRQLBSYHNZT46E3P7WU","download_json":"https://pith.science/pith/KBNXNJEHRQLBSYHNZT46E3P7WU.json","view_paper":"https://pith.science/paper/KBNXNJEH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2205.12674&json=true","fetch_graph":"https://pith.science/api/pith-number/KBNXNJEHRQLBSYHNZT46E3P7WU/graph.json","fetch_events":"https://pith.science/api/pith-number/KBNXNJEHRQLBSYHNZT46E3P7WU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/KBNXNJEHRQLBSYHNZT46E3P7WU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/KBNXNJEHRQLBSYHNZT46E3P7WU/action/storage_attestation","attest_author":"https://pith.science/pith/KBNXNJEHRQLBSYHNZT46E3P7WU/action/author_attestation","sign_citation":"https://pith.science/pith/KBNXNJEHRQLBSYHNZT46E3P7WU/action/citation_signature","submit_replication":"https://pith.science/pith/KBNXNJEHRQLBSYHNZT46E3P7WU/action/replication_record"}},"created_at":"2026-07-05T05:20:11.770754+00:00","updated_at":"2026-07-05T05:20:11.770754+00:00"}