{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:LHRIBK5ZDLHG7LZK7Q3625IP64","short_pith_number":"pith:LHRIBK5Z","schema_version":"1.0","canonical_sha256":"59e280abb91ace6faf2afc37ed750ff718627d1872a5905a7a2530a8652e6a08","source":{"kind":"arxiv","id":"2605.25451","version":1},"attestation_state":"computed","paper":{"title":"BigMac: Breaking the Pareto Frontier of Compute and Memory in Multimodal LLM Training","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Chao Jin, Chengxu Yang, Chenyu Wang, Shenglong Zhang, Tuo Dai, Xin Jin, Yufan Zhang, Yuhong Ge, Yuliang Liu, Zhouyang Li, Zili Zhang","submitted_at":"2026-05-25T06:01:45Z","abstract_excerpt":"Training multimodal large language models (MLLMs) is challenged by both model and data heterogeneity. Existing systems redesign the training pipeline to address these challenges, but remain bound by a Pareto frontier between compute and memory efficiency, improving one only at the expense of the other. We present BigMac, a new training pipeline for multimodal LLMs. The core idea of BigMac is to elegantly nest the encoder and generator computation into the original LLM pipeline, forming a dependency-safe nested pipeline structure. With this design, BigMac reduces the activation memory complexit"},"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":"2605.25451","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-05-25T06:01:45Z","cross_cats_sorted":[],"title_canon_sha256":"67968da5dfad49f03de281069cbe828fa2b1d288a1e0108ed626f7da9893306e","abstract_canon_sha256":"26fed20ecffba92c77157f0c1a3ab2ce5aee9f96925bd735350d232d6c2ca76b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-26T02:04:36.406791Z","signature_b64":"xmjZ6VxSz5+7gmuPb3BsoczwHod9ds8FYFWxHi29xB8u38zwX4xrAqKrOVc7DCOGz98zMCUwmWh9O971QQvDBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"59e280abb91ace6faf2afc37ed750ff718627d1872a5905a7a2530a8652e6a08","last_reissued_at":"2026-05-26T02:04:36.405922Z","signature_status":"signed_v1","first_computed_at":"2026-05-26T02:04:36.405922Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"BigMac: Breaking the Pareto Frontier of Compute and Memory in Multimodal LLM Training","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Chao Jin, Chengxu Yang, Chenyu Wang, Shenglong Zhang, Tuo Dai, Xin Jin, Yufan Zhang, Yuhong Ge, Yuliang Liu, Zhouyang Li, Zili Zhang","submitted_at":"2026-05-25T06:01:45Z","abstract_excerpt":"Training multimodal large language models (MLLMs) is challenged by both model and data heterogeneity. Existing systems redesign the training pipeline to address these challenges, but remain bound by a Pareto frontier between compute and memory efficiency, improving one only at the expense of the other. We present BigMac, a new training pipeline for multimodal LLMs. The core idea of BigMac is to elegantly nest the encoder and generator computation into the original LLM pipeline, forming a dependency-safe nested pipeline structure. With this design, BigMac reduces the activation memory complexit"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.25451","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/2605.25451/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":"2605.25451","created_at":"2026-05-26T02:04:36.406076+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.25451v1","created_at":"2026-05-26T02:04:36.406076+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.25451","created_at":"2026-05-26T02:04:36.406076+00:00"},{"alias_kind":"pith_short_12","alias_value":"LHRIBK5ZDLHG","created_at":"2026-05-26T02:04:36.406076+00:00"},{"alias_kind":"pith_short_16","alias_value":"LHRIBK5ZDLHG7LZK","created_at":"2026-05-26T02:04:36.406076+00:00"},{"alias_kind":"pith_short_8","alias_value":"LHRIBK5Z","created_at":"2026-05-26T02:04:36.406076+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/LHRIBK5ZDLHG7LZK7Q3625IP64","json":"https://pith.science/pith/LHRIBK5ZDLHG7LZK7Q3625IP64.json","graph_json":"https://pith.science/api/pith-number/LHRIBK5ZDLHG7LZK7Q3625IP64/graph.json","events_json":"https://pith.science/api/pith-number/LHRIBK5ZDLHG7LZK7Q3625IP64/events.json","paper":"https://pith.science/paper/LHRIBK5Z"},"agent_actions":{"view_html":"https://pith.science/pith/LHRIBK5ZDLHG7LZK7Q3625IP64","download_json":"https://pith.science/pith/LHRIBK5ZDLHG7LZK7Q3625IP64.json","view_paper":"https://pith.science/paper/LHRIBK5Z","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.25451&json=true","fetch_graph":"https://pith.science/api/pith-number/LHRIBK5ZDLHG7LZK7Q3625IP64/graph.json","fetch_events":"https://pith.science/api/pith-number/LHRIBK5ZDLHG7LZK7Q3625IP64/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/LHRIBK5ZDLHG7LZK7Q3625IP64/action/timestamp_anchor","attest_storage":"https://pith.science/pith/LHRIBK5ZDLHG7LZK7Q3625IP64/action/storage_attestation","attest_author":"https://pith.science/pith/LHRIBK5ZDLHG7LZK7Q3625IP64/action/author_attestation","sign_citation":"https://pith.science/pith/LHRIBK5ZDLHG7LZK7Q3625IP64/action/citation_signature","submit_replication":"https://pith.science/pith/LHRIBK5ZDLHG7LZK7Q3625IP64/action/replication_record"}},"created_at":"2026-05-26T02:04:36.406076+00:00","updated_at":"2026-05-26T02:04:36.406076+00:00"}