{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:SVKE3I5IXFVRWCDH4SUD33WVUJ","short_pith_number":"pith:SVKE3I5I","schema_version":"1.0","canonical_sha256":"95544da3a8b96b1b0867e4a83deed5a2460d50ba836832f105542ef249ed1032","source":{"kind":"arxiv","id":"2507.08771","version":2},"attestation_state":"computed","paper":{"title":"BlockFFN: Towards End-Side Acceleration-Friendly Mixture-of-Experts with Chunk-Level Activation Sparsity","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Chaojun Xiao, Chenyang Song, Maosong Sun, Weilin Zhao, Xu Han, Yingfa Chen, Yuxuan Li, Zhiyuan Liu","submitted_at":"2025-07-11T17:28:56Z","abstract_excerpt":"To alleviate the computational burden of large language models (LLMs), architectures with activation sparsity, represented by mixture-of-experts (MoE), have attracted increasing attention. However, the non-differentiable and inflexible routing of vanilla MoE hurts model performance. Moreover, while each token activates only a few parameters, these sparsely-activated architectures exhibit low chunk-level sparsity, indicating that the union of multiple consecutive tokens activates a large ratio of parameters. Such a sparsity pattern is unfriendly for acceleration under low-resource conditions (e"},"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":"2507.08771","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-07-11T17:28:56Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"d150a7be45c29a973edd986a6efcdc307b7fc7ecd19b59dd07d399a6687a14d2","abstract_canon_sha256":"015bd03ab890e3b424d2d200cc89a3474c4cad04111d609ca13eefb8db4fc4ba"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:45:37.423096Z","signature_b64":"WZ4K9JaggLJSqZ2LMtp6ng/cpJQ027M8Z3uSC8LEVhbkNWDpoPLxkAcPANm2XaENtsMZ1kHf4fyDZTtOU2VWAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"95544da3a8b96b1b0867e4a83deed5a2460d50ba836832f105542ef249ed1032","last_reissued_at":"2026-07-05T11:45:37.422600Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:45:37.422600Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"BlockFFN: Towards End-Side Acceleration-Friendly Mixture-of-Experts with Chunk-Level Activation Sparsity","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Chaojun Xiao, Chenyang Song, Maosong Sun, Weilin Zhao, Xu Han, Yingfa Chen, Yuxuan Li, Zhiyuan Liu","submitted_at":"2025-07-11T17:28:56Z","abstract_excerpt":"To alleviate the computational burden of large language models (LLMs), architectures with activation sparsity, represented by mixture-of-experts (MoE), have attracted increasing attention. However, the non-differentiable and inflexible routing of vanilla MoE hurts model performance. Moreover, while each token activates only a few parameters, these sparsely-activated architectures exhibit low chunk-level sparsity, indicating that the union of multiple consecutive tokens activates a large ratio of parameters. Such a sparsity pattern is unfriendly for acceleration under low-resource conditions (e"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.08771","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/2507.08771/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":"2507.08771","created_at":"2026-07-05T11:45:37.422656+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.08771v2","created_at":"2026-07-05T11:45:37.422656+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.08771","created_at":"2026-07-05T11:45:37.422656+00:00"},{"alias_kind":"pith_short_12","alias_value":"SVKE3I5IXFVR","created_at":"2026-07-05T11:45:37.422656+00:00"},{"alias_kind":"pith_short_16","alias_value":"SVKE3I5IXFVRWCDH","created_at":"2026-07-05T11:45:37.422656+00:00"},{"alias_kind":"pith_short_8","alias_value":"SVKE3I5I","created_at":"2026-07-05T11:45:37.422656+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.26287","citing_title":"GeMoE: Gating Entropy is All You Need for Uncertainty-aware Adaptive Routing in MoE-based Large Vision-Language Models","ref_index":51,"is_internal_anchor":false},{"citing_arxiv_id":"2605.30876","citing_title":"dMoE: dLLMs with Learnable Block Experts","ref_index":21,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SVKE3I5IXFVRWCDH4SUD33WVUJ","json":"https://pith.science/pith/SVKE3I5IXFVRWCDH4SUD33WVUJ.json","graph_json":"https://pith.science/api/pith-number/SVKE3I5IXFVRWCDH4SUD33WVUJ/graph.json","events_json":"https://pith.science/api/pith-number/SVKE3I5IXFVRWCDH4SUD33WVUJ/events.json","paper":"https://pith.science/paper/SVKE3I5I"},"agent_actions":{"view_html":"https://pith.science/pith/SVKE3I5IXFVRWCDH4SUD33WVUJ","download_json":"https://pith.science/pith/SVKE3I5IXFVRWCDH4SUD33WVUJ.json","view_paper":"https://pith.science/paper/SVKE3I5I","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.08771&json=true","fetch_graph":"https://pith.science/api/pith-number/SVKE3I5IXFVRWCDH4SUD33WVUJ/graph.json","fetch_events":"https://pith.science/api/pith-number/SVKE3I5IXFVRWCDH4SUD33WVUJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SVKE3I5IXFVRWCDH4SUD33WVUJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SVKE3I5IXFVRWCDH4SUD33WVUJ/action/storage_attestation","attest_author":"https://pith.science/pith/SVKE3I5IXFVRWCDH4SUD33WVUJ/action/author_attestation","sign_citation":"https://pith.science/pith/SVKE3I5IXFVRWCDH4SUD33WVUJ/action/citation_signature","submit_replication":"https://pith.science/pith/SVKE3I5IXFVRWCDH4SUD33WVUJ/action/replication_record"}},"created_at":"2026-07-05T11:45:37.422656+00:00","updated_at":"2026-07-05T11:45:37.422656+00:00"}