{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:GG7FYD47ZKO3LFCQZROI4COYS7","short_pith_number":"pith:GG7FYD47","schema_version":"1.0","canonical_sha256":"31be5c0f9fca9db59450cc5c8e09d897fd5377328a615ca058e5d51135ad4441","source":{"kind":"arxiv","id":"2502.12325","version":1},"attestation_state":"computed","paper":{"title":"From Dense to Dynamic: Token-Difficulty Driven MoEfication of Pre-Trained LLMs","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Devang Naik, Kumari Nishu, Mahyar Najibi, Maxwell Horton, Mehrdad Farajtabar, Minsik Cho, Moin Nabi, Sachin Mehta, Samira Abnar","submitted_at":"2025-02-17T21:12:57Z","abstract_excerpt":"Training large language models (LLMs) for different inference constraints is computationally expensive, limiting control over efficiency-accuracy trade-offs. Moreover, once trained, these models typically process tokens uniformly, regardless of their complexity, leading to static and inflexible behavior. In this paper, we introduce a post-training optimization framework, DynaMoE, that adapts a pre-trained dense LLM to a token-difficulty-driven Mixture-of-Experts model with minimal fine-tuning cost. This adaptation makes the model dynamic, with sensitivity control to customize the balance betwe"},"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":"2502.12325","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.CL","submitted_at":"2025-02-17T21:12:57Z","cross_cats_sorted":[],"title_canon_sha256":"a3fb1db05d509012c129c654c927052477246c39695b5ccb7ea44c1f37dad723","abstract_canon_sha256":"a82ffd82a0536972002cb9ca9654977ea4614509a446ff2d2f0b67c3e5a55c3d"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:15:53.639543Z","signature_b64":"uu60F3QnaPx9B+l13aqXiTVsxpQdlWzr5UdUmy7HcSqvgB/a45SehHqtCU/E68ILxFR+d2FsGveysdEn5FmuCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"31be5c0f9fca9db59450cc5c8e09d897fd5377328a615ca058e5d51135ad4441","last_reissued_at":"2026-07-05T10:15:53.639034Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:15:53.639034Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"From Dense to Dynamic: Token-Difficulty Driven MoEfication of Pre-Trained LLMs","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Devang Naik, Kumari Nishu, Mahyar Najibi, Maxwell Horton, Mehrdad Farajtabar, Minsik Cho, Moin Nabi, Sachin Mehta, Samira Abnar","submitted_at":"2025-02-17T21:12:57Z","abstract_excerpt":"Training large language models (LLMs) for different inference constraints is computationally expensive, limiting control over efficiency-accuracy trade-offs. Moreover, once trained, these models typically process tokens uniformly, regardless of their complexity, leading to static and inflexible behavior. In this paper, we introduce a post-training optimization framework, DynaMoE, that adapts a pre-trained dense LLM to a token-difficulty-driven Mixture-of-Experts model with minimal fine-tuning cost. This adaptation makes the model dynamic, with sensitivity control to customize the balance betwe"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.12325","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/2502.12325/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":"2502.12325","created_at":"2026-07-05T10:15:53.639090+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.12325v1","created_at":"2026-07-05T10:15:53.639090+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.12325","created_at":"2026-07-05T10:15:53.639090+00:00"},{"alias_kind":"pith_short_12","alias_value":"GG7FYD47ZKO3","created_at":"2026-07-05T10:15:53.639090+00:00"},{"alias_kind":"pith_short_16","alias_value":"GG7FYD47ZKO3LFCQ","created_at":"2026-07-05T10:15:53.639090+00:00"},{"alias_kind":"pith_short_8","alias_value":"GG7FYD47","created_at":"2026-07-05T10:15:53.639090+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2506.18887","citing_title":"Steering Conceptual Bias via Transformer Latent-Subspace Activation","ref_index":2023,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/GG7FYD47ZKO3LFCQZROI4COYS7","json":"https://pith.science/pith/GG7FYD47ZKO3LFCQZROI4COYS7.json","graph_json":"https://pith.science/api/pith-number/GG7FYD47ZKO3LFCQZROI4COYS7/graph.json","events_json":"https://pith.science/api/pith-number/GG7FYD47ZKO3LFCQZROI4COYS7/events.json","paper":"https://pith.science/paper/GG7FYD47"},"agent_actions":{"view_html":"https://pith.science/pith/GG7FYD47ZKO3LFCQZROI4COYS7","download_json":"https://pith.science/pith/GG7FYD47ZKO3LFCQZROI4COYS7.json","view_paper":"https://pith.science/paper/GG7FYD47","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.12325&json=true","fetch_graph":"https://pith.science/api/pith-number/GG7FYD47ZKO3LFCQZROI4COYS7/graph.json","fetch_events":"https://pith.science/api/pith-number/GG7FYD47ZKO3LFCQZROI4COYS7/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/GG7FYD47ZKO3LFCQZROI4COYS7/action/timestamp_anchor","attest_storage":"https://pith.science/pith/GG7FYD47ZKO3LFCQZROI4COYS7/action/storage_attestation","attest_author":"https://pith.science/pith/GG7FYD47ZKO3LFCQZROI4COYS7/action/author_attestation","sign_citation":"https://pith.science/pith/GG7FYD47ZKO3LFCQZROI4COYS7/action/citation_signature","submit_replication":"https://pith.science/pith/GG7FYD47ZKO3LFCQZROI4COYS7/action/replication_record"}},"created_at":"2026-07-05T10:15:53.639090+00:00","updated_at":"2026-07-05T10:15:53.639090+00:00"}