{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:T7DRUCYJ6FH25JNCP2X7WE4RNG","short_pith_number":"pith:T7DRUCYJ","canonical_record":{"source":{"id":"2605.05225","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2026-04-19T07:25:39Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"64eb6b04d158a0d934df44c311b8c21f4366ab8f4eeb683ee3f455c26532dcad","abstract_canon_sha256":"a427103785331e885adac95e4b24d14847dd5f9b1e9e9e29d86251fbb66dd751"},"schema_version":"1.0"},"canonical_sha256":"9fc71a0b09f14faea5a27eaffb139169a07f26d03e141432e871aed31622611a","source":{"kind":"arxiv","id":"2605.05225","version":3},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.05225","created_at":"2026-06-08T01:05:11Z"},{"alias_kind":"arxiv_version","alias_value":"2605.05225v3","created_at":"2026-06-08T01:05:11Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.05225","created_at":"2026-06-08T01:05:11Z"},{"alias_kind":"pith_short_12","alias_value":"T7DRUCYJ6FH2","created_at":"2026-06-08T01:05:11Z"},{"alias_kind":"pith_short_16","alias_value":"T7DRUCYJ6FH25JNC","created_at":"2026-06-08T01:05:11Z"},{"alias_kind":"pith_short_8","alias_value":"T7DRUCYJ","created_at":"2026-06-08T01:05:11Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:T7DRUCYJ6FH25JNCP2X7WE4RNG","target":"record","payload":{"canonical_record":{"source":{"id":"2605.05225","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2026-04-19T07:25:39Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"64eb6b04d158a0d934df44c311b8c21f4366ab8f4eeb683ee3f455c26532dcad","abstract_canon_sha256":"a427103785331e885adac95e4b24d14847dd5f9b1e9e9e29d86251fbb66dd751"},"schema_version":"1.0"},"canonical_sha256":"9fc71a0b09f14faea5a27eaffb139169a07f26d03e141432e871aed31622611a","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-08T01:05:11.581073Z","signature_b64":"HiLL47qcbfLoNRDNdghOvFYA+tvHuh8WMBZr6DH+VeYCkxWEhg7yUeR+CztTq6o+D2YXmBF+jO7nCNwSQPqpAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9fc71a0b09f14faea5a27eaffb139169a07f26d03e141432e871aed31622611a","last_reissued_at":"2026-06-08T01:05:11.580338Z","signature_status":"signed_v1","first_computed_at":"2026-06-08T01:05:11.580338Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.05225","source_version":3,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-08T01:05:11Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"wlHgDCe8N2pzC188pMLRWd9oCl6MiA6EuqqYrT+mCHMyrDkGFuYcpgnjSPdN9cr807901yxaW+56kRHDrLbnCQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-19T09:19:21.720298Z"},"content_sha256":"0e0f4351d9b49d90b29f62e11edf9f137463699808fb906e614c038baf1e9167","schema_version":"1.0","event_id":"sha256:0e0f4351d9b49d90b29f62e11edf9f137463699808fb906e614c038baf1e9167"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:T7DRUCYJ6FH25JNCP2X7WE4RNG","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"MACS: Modality-Aware Capacity Scaling for Efficient Multimodal MoE Inference","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"MACS scales expert capacity in multimodal MoE models using token entropy and real-time modality ratios to cut stragglers.","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Bo Li, Chuan Wu, Shaolin Zhu","submitted_at":"2026-04-19T07:25:39Z","abstract_excerpt":"Mixture-of-Experts Multimodal Large Language Models (MoE MLLMs) suffer from a significant efficiency bottleneck during Expert Parallelism (EP) inference due to the straggler effect. This issue is worsened in the multimodal context, as existing token-count-based load balancing methods fail to address two unique challenges: (1) Information Heterogeneity, where numerous redundant visual tokens are treated equally to semantically critical ones, and (2) Modality Dynamics, where varying visual to text ratios across tasks lead to resource misallocation. To address these challenges, we propose MACS (M"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"MACS significantly outperforms existing methods on various multimodal benchmarks, providing a novel and robust solution for the efficient deployment of MoE MLLMs in EP inference.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the Entropy-Weighted Load mechanism and Dynamic Modality-Adaptive Capacity mechanism will reliably reduce the straggler effect across diverse multimodal inputs and hardware setups without any training or task-specific tuning.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"MACS improves inference speed in multimodal MoE models by entropy-weighted balancing of visual tokens and real-time modality-adaptive expert capacity allocation.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"MACS scales expert capacity in multimodal MoE models using token entropy and real-time modality ratios to cut stragglers.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"182cda16183ab6be680baf386805d482b9b46f0450cd67ca7598271b29dc54e0"},"source":{"id":"2605.05225","kind":"arxiv","version":3},"verdict":{"id":"f394d361-3819-4d03-bacd-836a034f1401","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-11T01:26:15.758356Z","strongest_claim":"MACS significantly outperforms existing methods on various multimodal benchmarks, providing a novel and robust solution for the efficient deployment of MoE MLLMs in EP inference.","one_line_summary":"MACS improves inference speed in multimodal MoE models by entropy-weighted balancing of visual tokens and real-time modality-adaptive expert capacity allocation.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the Entropy-Weighted Load mechanism and Dynamic Modality-Adaptive Capacity mechanism will reliably reduce the straggler effect across diverse multimodal inputs and hardware setups without any training or task-specific tuning.","pith_extraction_headline":"MACS scales expert capacity in multimodal MoE models using token entropy and real-time modality ratios to cut stragglers."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.05225/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":2,"snapshot_sha256":"cd57a87d09a04b700fcd6acebceeec2d31d97653f32e0ac7eb4c96a4949c2cc1"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"f394d361-3819-4d03-bacd-836a034f1401"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-06-08T01:05:11Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"JR2U3Eef2LCU0lecrgXq9RJzqIOHbORMtbgMsmI064ugjVkQF0aXY1LvGKG3uWiPeTKGCgw0ZuOLtyxbFBZLBQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-19T09:19:21.721407Z"},"content_sha256":"91b4885bb5fd9816c026ffccb09149e270b48ad1e957385489d050765ada6c6d","schema_version":"1.0","event_id":"sha256:91b4885bb5fd9816c026ffccb09149e270b48ad1e957385489d050765ada6c6d"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/T7DRUCYJ6FH25JNCP2X7WE4RNG/bundle.json","state_url":"https://pith.science/pith/T7DRUCYJ6FH25JNCP2X7WE4RNG/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/T7DRUCYJ6FH25JNCP2X7WE4RNG/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-08-19T09:19:21Z","links":{"resolver":"https://pith.science/pith/T7DRUCYJ6FH25JNCP2X7WE4RNG","bundle":"https://pith.science/pith/T7DRUCYJ6FH25JNCP2X7WE4RNG/bundle.json","state":"https://pith.science/pith/T7DRUCYJ6FH25JNCP2X7WE4RNG/state.json","well_known_bundle":"https://pith.science/.well-known/pith/T7DRUCYJ6FH25JNCP2X7WE4RNG/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:T7DRUCYJ6FH25JNCP2X7WE4RNG","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"a427103785331e885adac95e4b24d14847dd5f9b1e9e9e29d86251fbb66dd751","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2026-04-19T07:25:39Z","title_canon_sha256":"64eb6b04d158a0d934df44c311b8c21f4366ab8f4eeb683ee3f455c26532dcad"},"schema_version":"1.0","source":{"id":"2605.05225","kind":"arxiv","version":3}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.05225","created_at":"2026-06-08T01:05:11Z"},{"alias_kind":"arxiv_version","alias_value":"2605.05225v3","created_at":"2026-06-08T01:05:11Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.05225","created_at":"2026-06-08T01:05:11Z"},{"alias_kind":"pith_short_12","alias_value":"T7DRUCYJ6FH2","created_at":"2026-06-08T01:05:11Z"},{"alias_kind":"pith_short_16","alias_value":"T7DRUCYJ6FH25JNC","created_at":"2026-06-08T01:05:11Z"},{"alias_kind":"pith_short_8","alias_value":"T7DRUCYJ","created_at":"2026-06-08T01:05:11Z"}],"graph_snapshots":[{"event_id":"sha256:91b4885bb5fd9816c026ffccb09149e270b48ad1e957385489d050765ada6c6d","target":"graph","created_at":"2026-06-08T01:05:11Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"MACS significantly outperforms existing methods on various multimodal benchmarks, providing a novel and robust solution for the efficient deployment of MoE MLLMs in EP inference."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the Entropy-Weighted Load mechanism and Dynamic Modality-Adaptive Capacity mechanism will reliably reduce the straggler effect across diverse multimodal inputs and hardware setups without any training or task-specific tuning."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"MACS improves inference speed in multimodal MoE models by entropy-weighted balancing of visual tokens and real-time modality-adaptive expert capacity allocation."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"MACS scales expert capacity in multimodal MoE models using token entropy and real-time modality ratios to cut stragglers."}],"snapshot_sha256":"182cda16183ab6be680baf386805d482b9b46f0450cd67ca7598271b29dc54e0"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"cd57a87d09a04b700fcd6acebceeec2d31d97653f32e0ac7eb4c96a4949c2cc1"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2605.05225/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Mixture-of-Experts Multimodal Large Language Models (MoE MLLMs) suffer from a significant efficiency bottleneck during Expert Parallelism (EP) inference due to the straggler effect. This issue is worsened in the multimodal context, as existing token-count-based load balancing methods fail to address two unique challenges: (1) Information Heterogeneity, where numerous redundant visual tokens are treated equally to semantically critical ones, and (2) Modality Dynamics, where varying visual to text ratios across tasks lead to resource misallocation. To address these challenges, we propose MACS (M","authors_text":"Bo Li, Chuan Wu, Shaolin Zhu","cross_cats":["cs.AI"],"headline":"MACS scales expert capacity in multimodal MoE models using token entropy and real-time modality ratios to cut stragglers.","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2026-04-19T07:25:39Z","title":"MACS: Modality-Aware Capacity Scaling for Efficient Multimodal MoE Inference"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.05225","kind":"arxiv","version":3},"verdict":{"created_at":"2026-05-11T01:26:15.758356Z","id":"f394d361-3819-4d03-bacd-836a034f1401","model_set":{"reader":"grok-4.3"},"one_line_summary":"MACS improves inference speed in multimodal MoE models by entropy-weighted balancing of visual tokens and real-time modality-adaptive expert capacity allocation.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"MACS scales expert capacity in multimodal MoE models using token entropy and real-time modality ratios to cut stragglers.","strongest_claim":"MACS significantly outperforms existing methods on various multimodal benchmarks, providing a novel and robust solution for the efficient deployment of MoE MLLMs in EP inference.","weakest_assumption":"That the Entropy-Weighted Load mechanism and Dynamic Modality-Adaptive Capacity mechanism will reliably reduce the straggler effect across diverse multimodal inputs and hardware setups without any training or task-specific tuning."}},"verdict_id":"f394d361-3819-4d03-bacd-836a034f1401"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:0e0f4351d9b49d90b29f62e11edf9f137463699808fb906e614c038baf1e9167","target":"record","created_at":"2026-06-08T01:05:11Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"a427103785331e885adac95e4b24d14847dd5f9b1e9e9e29d86251fbb66dd751","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.LG","submitted_at":"2026-04-19T07:25:39Z","title_canon_sha256":"64eb6b04d158a0d934df44c311b8c21f4366ab8f4eeb683ee3f455c26532dcad"},"schema_version":"1.0","source":{"id":"2605.05225","kind":"arxiv","version":3}},"canonical_sha256":"9fc71a0b09f14faea5a27eaffb139169a07f26d03e141432e871aed31622611a","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"9fc71a0b09f14faea5a27eaffb139169a07f26d03e141432e871aed31622611a","first_computed_at":"2026-06-08T01:05:11.580338Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-06-08T01:05:11.580338Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"HiLL47qcbfLoNRDNdghOvFYA+tvHuh8WMBZr6DH+VeYCkxWEhg7yUeR+CztTq6o+D2YXmBF+jO7nCNwSQPqpAQ==","signature_status":"signed_v1","signed_at":"2026-06-08T01:05:11.581073Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.05225","source_kind":"arxiv","source_version":3}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:0e0f4351d9b49d90b29f62e11edf9f137463699808fb906e614c038baf1e9167","sha256:91b4885bb5fd9816c026ffccb09149e270b48ad1e957385489d050765ada6c6d"],"state_sha256":"0ac4082566c2070649ec8224ddda67ae452cf416e00a30666d1fdf791a93d2ef"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"c+KwSXoNGlVMqWA5mSiCkrcehQ2x4XY1QQiWSINJTAlUTHcEapDWuC1DL6ZY33GkGlKCjUT9GIGDfk+/2e7eCg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-19T09:19:21.726111Z","bundle_sha256":"8019f4c8cdbd92a0b9a762ebf14e9602850e4013cde6319a525b3ff4e3b8a463"}}