{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:RRSTML4DVVYVJZKYGQ5DOUHFZT","short_pith_number":"pith:RRSTML4D","schema_version":"1.0","canonical_sha256":"8c65362f83ad7154e558343a3750e5ccdb728b63dfb0e76c80cfbb8b432b741c","source":{"kind":"arxiv","id":"2508.05305","version":2},"attestation_state":"computed","paper":{"title":"SONAR-LLM: Autoregressive Transformer that Thinks in Sentence Embeddings and Speaks in Tokens","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Andrey Kuznetsov, Anna Borisiuk, Anton Razzhigaev, Aysel Mirzoeva, Elizaveta Goncharova, Nikita Dragunov, Nikita Kurdiukov, Temurbek Rahmatullaev","submitted_at":"2025-08-07T12:03:44Z","abstract_excerpt":"The recently proposed Large Concept Model (LCM) generates text by predicting a sequence of sentence-level embeddings and training with either mean-squared error or diffusion objectives. We present SONAR-LLM, a decoder-only transformer that \"thinks\" in the same continuous SONAR embedding space, yet is supervised through token-level cross-entropy propagated via the frozen SONAR decoder. This hybrid objective retains the semantic abstraction of LCM while eliminating its diffusion sampler and restoring a likelihood-based training signal. Across model sizes from 39M to 1.3B parameters, SONAR-LLM at"},"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":"2508.05305","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-08-07T12:03:44Z","cross_cats_sorted":[],"title_canon_sha256":"de4e36a94395f16b65a7a4af48102170db92d4a8e5dcb323d13c415af313d478","abstract_canon_sha256":"ddde7bbd7eeb9ea9aabaee42b15b57b9453fad3e844941daf0d7a8c9303ebdfe"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-27T01:04:49.807306Z","signature_b64":"XtgYBPCASD2nYjr1+ephS/cDcJBMeq1Gwq9ZAhuM/MU0rmuQZdGtmkQB34M+fdo/2SyAIdJf8XIv5lLx182iCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8c65362f83ad7154e558343a3750e5ccdb728b63dfb0e76c80cfbb8b432b741c","last_reissued_at":"2026-05-27T01:04:49.806590Z","signature_status":"signed_v1","first_computed_at":"2026-05-27T01:04:49.806590Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"SONAR-LLM: Autoregressive Transformer that Thinks in Sentence Embeddings and Speaks in Tokens","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Andrey Kuznetsov, Anna Borisiuk, Anton Razzhigaev, Aysel Mirzoeva, Elizaveta Goncharova, Nikita Dragunov, Nikita Kurdiukov, Temurbek Rahmatullaev","submitted_at":"2025-08-07T12:03:44Z","abstract_excerpt":"The recently proposed Large Concept Model (LCM) generates text by predicting a sequence of sentence-level embeddings and training with either mean-squared error or diffusion objectives. We present SONAR-LLM, a decoder-only transformer that \"thinks\" in the same continuous SONAR embedding space, yet is supervised through token-level cross-entropy propagated via the frozen SONAR decoder. This hybrid objective retains the semantic abstraction of LCM while eliminating its diffusion sampler and restoring a likelihood-based training signal. Across model sizes from 39M to 1.3B parameters, SONAR-LLM at"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.05305","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/2508.05305/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":"2508.05305","created_at":"2026-05-27T01:04:49.806673+00:00"},{"alias_kind":"arxiv_version","alias_value":"2508.05305v2","created_at":"2026-05-27T01:04:49.806673+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.05305","created_at":"2026-05-27T01:04:49.806673+00:00"},{"alias_kind":"pith_short_12","alias_value":"RRSTML4DVVYV","created_at":"2026-05-27T01:04:49.806673+00:00"},{"alias_kind":"pith_short_16","alias_value":"RRSTML4DVVYVJZKY","created_at":"2026-05-27T01:04:49.806673+00:00"},{"alias_kind":"pith_short_8","alias_value":"RRSTML4D","created_at":"2026-05-27T01:04:49.806673+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2605.20194","citing_title":"Parallel LLM Reasoning for Bias-Resilient, Robust Conceptual Abstraction","ref_index":13,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/RRSTML4DVVYVJZKYGQ5DOUHFZT","json":"https://pith.science/pith/RRSTML4DVVYVJZKYGQ5DOUHFZT.json","graph_json":"https://pith.science/api/pith-number/RRSTML4DVVYVJZKYGQ5DOUHFZT/graph.json","events_json":"https://pith.science/api/pith-number/RRSTML4DVVYVJZKYGQ5DOUHFZT/events.json","paper":"https://pith.science/paper/RRSTML4D"},"agent_actions":{"view_html":"https://pith.science/pith/RRSTML4DVVYVJZKYGQ5DOUHFZT","download_json":"https://pith.science/pith/RRSTML4DVVYVJZKYGQ5DOUHFZT.json","view_paper":"https://pith.science/paper/RRSTML4D","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2508.05305&json=true","fetch_graph":"https://pith.science/api/pith-number/RRSTML4DVVYVJZKYGQ5DOUHFZT/graph.json","fetch_events":"https://pith.science/api/pith-number/RRSTML4DVVYVJZKYGQ5DOUHFZT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/RRSTML4DVVYVJZKYGQ5DOUHFZT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/RRSTML4DVVYVJZKYGQ5DOUHFZT/action/storage_attestation","attest_author":"https://pith.science/pith/RRSTML4DVVYVJZKYGQ5DOUHFZT/action/author_attestation","sign_citation":"https://pith.science/pith/RRSTML4DVVYVJZKYGQ5DOUHFZT/action/citation_signature","submit_replication":"https://pith.science/pith/RRSTML4DVVYVJZKYGQ5DOUHFZT/action/replication_record"}},"created_at":"2026-05-27T01:04:49.806673+00:00","updated_at":"2026-05-27T01:04:49.806673+00:00"}