{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:JOKFXRWRD5W75GZNK2XESXZCQP","short_pith_number":"pith:JOKFXRWR","schema_version":"1.0","canonical_sha256":"4b945bc6d11f6dfe9b2d56ae495f2283cb0bdaa7b4dba0cef2c16705557836cd","source":{"kind":"arxiv","id":"2503.06881","version":1},"attestation_state":"computed","paper":{"title":"ResMoE: Space-efficient Compression of Mixture of Experts LLMs via Residual Restoration","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Bita Darvish Rouhani, Girish Varatkar, Hanghang Tong, Jingrui He, Mengting Ai, Ritchie Zhao, Tianxin Wei, Xianfeng Tang, Yifan Chen, Zhichen Zeng","submitted_at":"2025-03-10T03:15:54Z","abstract_excerpt":"Mixture-of-Experts (MoE) Transformer, the backbone architecture of multiple phenomenal language models, leverages sparsity by activating only a fraction of model parameters for each input token. The sparse structure, while allowing constant time costs, results in space inefficiency: we still need to load all the model parameters during inference. We introduce ResMoE, an innovative MoE approximation framework that utilizes Wasserstein barycenter to extract a common expert (barycenter expert) and approximate the residuals between this barycenter expert and the original ones. ResMoE enhances the "},"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":"2503.06881","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-03-10T03:15:54Z","cross_cats_sorted":[],"title_canon_sha256":"9f99faaec5d1e84b8fcd82ba74206e80951a69577d1c204f1c88cfc5be7f4692","abstract_canon_sha256":"481480057dfec05d3971e127333f2a0658a4ef79d923b18af5f87c94c4b37e66"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:27:45.052883Z","signature_b64":"J4Rik7EblYXNdsxKQ507Z5Le+uR8EK8V3o4dMQroWPbGm6aQ+HB/AidYo8Hjyre+aHrGMWmPSecRemXn+cD+CQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4b945bc6d11f6dfe9b2d56ae495f2283cb0bdaa7b4dba0cef2c16705557836cd","last_reissued_at":"2026-07-05T10:27:45.051706Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:27:45.051706Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"ResMoE: Space-efficient Compression of Mixture of Experts LLMs via Residual Restoration","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Bita Darvish Rouhani, Girish Varatkar, Hanghang Tong, Jingrui He, Mengting Ai, Ritchie Zhao, Tianxin Wei, Xianfeng Tang, Yifan Chen, Zhichen Zeng","submitted_at":"2025-03-10T03:15:54Z","abstract_excerpt":"Mixture-of-Experts (MoE) Transformer, the backbone architecture of multiple phenomenal language models, leverages sparsity by activating only a fraction of model parameters for each input token. The sparse structure, while allowing constant time costs, results in space inefficiency: we still need to load all the model parameters during inference. We introduce ResMoE, an innovative MoE approximation framework that utilizes Wasserstein barycenter to extract a common expert (barycenter expert) and approximate the residuals between this barycenter expert and the original ones. ResMoE enhances the "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2503.06881","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/2503.06881/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":"2503.06881","created_at":"2026-07-05T10:27:45.051867+00:00"},{"alias_kind":"arxiv_version","alias_value":"2503.06881v1","created_at":"2026-07-05T10:27:45.051867+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2503.06881","created_at":"2026-07-05T10:27:45.051867+00:00"},{"alias_kind":"pith_short_12","alias_value":"JOKFXRWRD5W7","created_at":"2026-07-05T10:27:45.051867+00:00"},{"alias_kind":"pith_short_16","alias_value":"JOKFXRWRD5W75GZN","created_at":"2026-07-05T10:27:45.051867+00:00"},{"alias_kind":"pith_short_8","alias_value":"JOKFXRWR","created_at":"2026-07-05T10:27:45.051867+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/JOKFXRWRD5W75GZNK2XESXZCQP","json":"https://pith.science/pith/JOKFXRWRD5W75GZNK2XESXZCQP.json","graph_json":"https://pith.science/api/pith-number/JOKFXRWRD5W75GZNK2XESXZCQP/graph.json","events_json":"https://pith.science/api/pith-number/JOKFXRWRD5W75GZNK2XESXZCQP/events.json","paper":"https://pith.science/paper/JOKFXRWR"},"agent_actions":{"view_html":"https://pith.science/pith/JOKFXRWRD5W75GZNK2XESXZCQP","download_json":"https://pith.science/pith/JOKFXRWRD5W75GZNK2XESXZCQP.json","view_paper":"https://pith.science/paper/JOKFXRWR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2503.06881&json=true","fetch_graph":"https://pith.science/api/pith-number/JOKFXRWRD5W75GZNK2XESXZCQP/graph.json","fetch_events":"https://pith.science/api/pith-number/JOKFXRWRD5W75GZNK2XESXZCQP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/JOKFXRWRD5W75GZNK2XESXZCQP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/JOKFXRWRD5W75GZNK2XESXZCQP/action/storage_attestation","attest_author":"https://pith.science/pith/JOKFXRWRD5W75GZNK2XESXZCQP/action/author_attestation","sign_citation":"https://pith.science/pith/JOKFXRWRD5W75GZNK2XESXZCQP/action/citation_signature","submit_replication":"https://pith.science/pith/JOKFXRWRD5W75GZNK2XESXZCQP/action/replication_record"}},"created_at":"2026-07-05T10:27:45.051867+00:00","updated_at":"2026-07-05T10:27:45.051867+00:00"}