{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:KCW2L5XH64ZHKHS34OAHZHYVLP","short_pith_number":"pith:KCW2L5XH","schema_version":"1.0","canonical_sha256":"50ada5f6e7f732751e5be3807c9f155bdcae91ed03bbbfd2dc50854592ec6a4c","source":{"kind":"arxiv","id":"2410.10814","version":2},"attestation_state":"computed","paper":{"title":"Your Mixture-of-Experts LLM Is Secretly an Embedding Model For Free","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Tianyi Zhou, Ziyue Li","submitted_at":"2024-10-14T17:59:44Z","abstract_excerpt":"While large language models (LLMs) excel on generation tasks, their decoder-only architecture often limits their potential as embedding models if no further representation finetuning is applied. Does this contradict their claim of generalists? To answer the question, we take a closer look at Mixture-of-Experts (MoE) LLMs. Our study shows that the expert routers in MoE LLMs can serve as an off-the-shelf embedding model with promising performance on a diverse class of embedding-focused tasks, without requiring any finetuning. Moreover, our extensive analysis shows that the MoE routing weights (R"},"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":"2410.10814","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-10-14T17:59:44Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"011b7bc19332d85ba5470f8e4bc155819f2b2c9c17c030d2d0d5decb264f6569","abstract_canon_sha256":"b86ba7932905b6b3f44bd0d9396b0adfa132e4b8935b96bde262259dfe36360e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:21:08.552539Z","signature_b64":"3DHQN/FbCuGTrWqUBKlvzpKroGIkIxcdbquz/KwL/kna0uu/g6oMxXwr97/JfBDPcM7Mizx71sQcfR0hNiGNDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"50ada5f6e7f732751e5be3807c9f155bdcae91ed03bbbfd2dc50854592ec6a4c","last_reissued_at":"2026-07-05T09:21:08.552075Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:21:08.552075Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Your Mixture-of-Experts LLM Is Secretly an Embedding Model For Free","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Tianyi Zhou, Ziyue Li","submitted_at":"2024-10-14T17:59:44Z","abstract_excerpt":"While large language models (LLMs) excel on generation tasks, their decoder-only architecture often limits their potential as embedding models if no further representation finetuning is applied. Does this contradict their claim of generalists? To answer the question, we take a closer look at Mixture-of-Experts (MoE) LLMs. Our study shows that the expert routers in MoE LLMs can serve as an off-the-shelf embedding model with promising performance on a diverse class of embedding-focused tasks, without requiring any finetuning. Moreover, our extensive analysis shows that the MoE routing weights (R"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.10814","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/2410.10814/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":"2410.10814","created_at":"2026-07-05T09:21:08.552130+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.10814v2","created_at":"2026-07-05T09:21:08.552130+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.10814","created_at":"2026-07-05T09:21:08.552130+00:00"},{"alias_kind":"pith_short_12","alias_value":"KCW2L5XH64ZH","created_at":"2026-07-05T09:21:08.552130+00:00"},{"alias_kind":"pith_short_16","alias_value":"KCW2L5XH64ZHKHS3","created_at":"2026-07-05T09:21:08.552130+00:00"},{"alias_kind":"pith_short_8","alias_value":"KCW2L5XH","created_at":"2026-07-05T09:21:08.552130+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.27069","citing_title":"Towards Explainable Adjudicative Variance: Quantifying Judicial Discretion via Gated Multi-Task Learning","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2606.27069","citing_title":"Towards Explainable Adjudicative Variance: Quantifying Judicial Discretion via Gated Multi-Task Learning","ref_index":9,"is_internal_anchor":false},{"citing_arxiv_id":"2509.24621","citing_title":"FreeRet: MLLMs as Training-Free Retrievers","ref_index":14,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/KCW2L5XH64ZHKHS34OAHZHYVLP","json":"https://pith.science/pith/KCW2L5XH64ZHKHS34OAHZHYVLP.json","graph_json":"https://pith.science/api/pith-number/KCW2L5XH64ZHKHS34OAHZHYVLP/graph.json","events_json":"https://pith.science/api/pith-number/KCW2L5XH64ZHKHS34OAHZHYVLP/events.json","paper":"https://pith.science/paper/KCW2L5XH"},"agent_actions":{"view_html":"https://pith.science/pith/KCW2L5XH64ZHKHS34OAHZHYVLP","download_json":"https://pith.science/pith/KCW2L5XH64ZHKHS34OAHZHYVLP.json","view_paper":"https://pith.science/paper/KCW2L5XH","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.10814&json=true","fetch_graph":"https://pith.science/api/pith-number/KCW2L5XH64ZHKHS34OAHZHYVLP/graph.json","fetch_events":"https://pith.science/api/pith-number/KCW2L5XH64ZHKHS34OAHZHYVLP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/KCW2L5XH64ZHKHS34OAHZHYVLP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/KCW2L5XH64ZHKHS34OAHZHYVLP/action/storage_attestation","attest_author":"https://pith.science/pith/KCW2L5XH64ZHKHS34OAHZHYVLP/action/author_attestation","sign_citation":"https://pith.science/pith/KCW2L5XH64ZHKHS34OAHZHYVLP/action/citation_signature","submit_replication":"https://pith.science/pith/KCW2L5XH64ZHKHS34OAHZHYVLP/action/replication_record"}},"created_at":"2026-07-05T09:21:08.552130+00:00","updated_at":"2026-07-05T09:21:08.552130+00:00"}