{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:QRP2LTGWBMLFRRL7OQRI6UKTK4","short_pith_number":"pith:QRP2LTGW","schema_version":"1.0","canonical_sha256":"845fa5ccd60b1658c57f74228f5153570574c21a2123b430807f8dee3bbe4994","source":{"kind":"arxiv","id":"2505.12629","version":1},"attestation_state":"computed","paper":{"title":"Enhancing Latent Computation in Transformers with Latent Tokens","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Bolin Ding, Yaliang Li, Yanxi Chen, Yuchang Sun","submitted_at":"2025-05-19T02:35:53Z","abstract_excerpt":"Augmenting large language models (LLMs) with auxiliary tokens has emerged as a promising strategy for enhancing model performance. In this work, we introduce a lightweight method termed latent tokens; these are dummy tokens that may be non-interpretable in natural language but steer the autoregressive decoding process of a Transformer-based LLM via the attention mechanism. The proposed latent tokens can be seamlessly integrated with a pre-trained Transformer, trained in a parameter-efficient manner, and applied flexibly at inference time, while adding minimal complexity overhead to the existin"},"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":"2505.12629","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2025-05-19T02:35:53Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"03debe93cb6ddcd3bf6f51b870e51bb7ead10a3ff6313e77689587e634e4bb89","abstract_canon_sha256":"bcad76b6589af12dfdcddbc5f6ca8455e51b1d0d7e34ba5ebc3fee9dfe5fcfcc"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:05:07.264899Z","signature_b64":"E8PQwf7XC+sBsHVeOl9pj3i4kvNckzs2HKer53vIHhr9FhF3D+fWsesBylRY00c2DpD/GULPPLE8M2e0Mc/qDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"845fa5ccd60b1658c57f74228f5153570574c21a2123b430807f8dee3bbe4994","last_reissued_at":"2026-07-05T11:05:07.264460Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:05:07.264460Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Enhancing Latent Computation in Transformers with Latent Tokens","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Bolin Ding, Yaliang Li, Yanxi Chen, Yuchang Sun","submitted_at":"2025-05-19T02:35:53Z","abstract_excerpt":"Augmenting large language models (LLMs) with auxiliary tokens has emerged as a promising strategy for enhancing model performance. In this work, we introduce a lightweight method termed latent tokens; these are dummy tokens that may be non-interpretable in natural language but steer the autoregressive decoding process of a Transformer-based LLM via the attention mechanism. The proposed latent tokens can be seamlessly integrated with a pre-trained Transformer, trained in a parameter-efficient manner, and applied flexibly at inference time, while adding minimal complexity overhead to the existin"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.12629","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/2505.12629/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":"2505.12629","created_at":"2026-07-05T11:05:07.264523+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.12629v1","created_at":"2026-07-05T11:05:07.264523+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.12629","created_at":"2026-07-05T11:05:07.264523+00:00"},{"alias_kind":"pith_short_12","alias_value":"QRP2LTGWBMLF","created_at":"2026-07-05T11:05:07.264523+00:00"},{"alias_kind":"pith_short_16","alias_value":"QRP2LTGWBMLFRRL7","created_at":"2026-07-05T11:05:07.264523+00:00"},{"alias_kind":"pith_short_8","alias_value":"QRP2LTGW","created_at":"2026-07-05T11:05:07.264523+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":4,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.18089","citing_title":"From Reasoning Traces to Reusable Modules: Understanding Compositional Generalization in Language Model Reasoning","ref_index":50,"is_internal_anchor":false},{"citing_arxiv_id":"2605.30343","citing_title":"Unlocking the Working Memory of Large Language Models for Latent Reasoning","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09346","citing_title":"RuPLaR : Efficient Latent Compression of LLM Reasoning Chains with Rule-Based Priors From Multi-Step to One-Step","ref_index":14,"is_internal_anchor":false},{"citing_arxiv_id":"2604.21027","citing_title":"HypEHR: Hyperbolic Modeling of Electronic Health Records for Efficient Question Answering","ref_index":88,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/QRP2LTGWBMLFRRL7OQRI6UKTK4","json":"https://pith.science/pith/QRP2LTGWBMLFRRL7OQRI6UKTK4.json","graph_json":"https://pith.science/api/pith-number/QRP2LTGWBMLFRRL7OQRI6UKTK4/graph.json","events_json":"https://pith.science/api/pith-number/QRP2LTGWBMLFRRL7OQRI6UKTK4/events.json","paper":"https://pith.science/paper/QRP2LTGW"},"agent_actions":{"view_html":"https://pith.science/pith/QRP2LTGWBMLFRRL7OQRI6UKTK4","download_json":"https://pith.science/pith/QRP2LTGWBMLFRRL7OQRI6UKTK4.json","view_paper":"https://pith.science/paper/QRP2LTGW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.12629&json=true","fetch_graph":"https://pith.science/api/pith-number/QRP2LTGWBMLFRRL7OQRI6UKTK4/graph.json","fetch_events":"https://pith.science/api/pith-number/QRP2LTGWBMLFRRL7OQRI6UKTK4/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/QRP2LTGWBMLFRRL7OQRI6UKTK4/action/timestamp_anchor","attest_storage":"https://pith.science/pith/QRP2LTGWBMLFRRL7OQRI6UKTK4/action/storage_attestation","attest_author":"https://pith.science/pith/QRP2LTGWBMLFRRL7OQRI6UKTK4/action/author_attestation","sign_citation":"https://pith.science/pith/QRP2LTGWBMLFRRL7OQRI6UKTK4/action/citation_signature","submit_replication":"https://pith.science/pith/QRP2LTGWBMLFRRL7OQRI6UKTK4/action/replication_record"}},"created_at":"2026-07-05T11:05:07.264523+00:00","updated_at":"2026-07-05T11:05:07.264523+00:00"}