{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:JSC4J7ABFLU4NBQSAHNZZZZPTT","short_pith_number":"pith:JSC4J7AB","schema_version":"1.0","canonical_sha256":"4c85c4fc012ae9c6861201db9ce72f9cc2e670d58dce20e76fba9264fbc14c2c","source":{"kind":"arxiv","id":"2605.27678","version":1},"attestation_state":"computed","paper":{"title":"Heterogeneous Parallelism for Multimodal Large Language Model Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.DC"],"primary_cat":"cs.LG","authors_text":"Akash Mehra, Ali Roshan Ghias, Ashwath Aithal, Eric Harper, Hao Wu, Kamran Jafari, Li Ding, Nima Tajbakhsh, Parth Mannan, Pranav Prashant Thombre, Shifang Xu, Yashaswi Karnati, Yu Yao","submitted_at":"2026-05-26T20:53:06Z","abstract_excerpt":"Foundation model training is becoming multimodal, from post-training pipelines to large-scale pretraining. As modality coverage broadens, context windows grow, and encoder LLM scales diverge, a single LLM-centric TP/CP/PP/DP/EP layout increasingly limits throughput. This coupling forces encoders to inherit LLM-driven sharding and placement choices that can add communication, limit encoder parallelism, or constrain the LLM schedule; the mismatch is most pronounced at long contexts, where LLM context parallelism is needed for the fused multimodal sequence but encoder inputs remain bounded. We pr"},"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":"2605.27678","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-26T20:53:06Z","cross_cats_sorted":["cs.DC"],"title_canon_sha256":"e40d1ff09845487f4d2dacfea5b219eaecc4751b06ef6e00b5830fb78562b92a","abstract_canon_sha256":"5c782e6dfcf6c410651bd0804b2a36d85015b41b090e72f934a835a06dcbb896"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-28T01:04:45.992004Z","signature_b64":"9STNxoImCHpYOg5+trEy2pZLGm2ur6tajtnOMsA/mU3epd/6KZYvMg1u5cJjcp4mP5XhT54xzgknTlu5FPJOAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4c85c4fc012ae9c6861201db9ce72f9cc2e670d58dce20e76fba9264fbc14c2c","last_reissued_at":"2026-05-28T01:04:45.991612Z","signature_status":"signed_v1","first_computed_at":"2026-05-28T01:04:45.991612Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Heterogeneous Parallelism for Multimodal Large Language Model Training","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.DC"],"primary_cat":"cs.LG","authors_text":"Akash Mehra, Ali Roshan Ghias, Ashwath Aithal, Eric Harper, Hao Wu, Kamran Jafari, Li Ding, Nima Tajbakhsh, Parth Mannan, Pranav Prashant Thombre, Shifang Xu, Yashaswi Karnati, Yu Yao","submitted_at":"2026-05-26T20:53:06Z","abstract_excerpt":"Foundation model training is becoming multimodal, from post-training pipelines to large-scale pretraining. As modality coverage broadens, context windows grow, and encoder LLM scales diverge, a single LLM-centric TP/CP/PP/DP/EP layout increasingly limits throughput. This coupling forces encoders to inherit LLM-driven sharding and placement choices that can add communication, limit encoder parallelism, or constrain the LLM schedule; the mismatch is most pronounced at long contexts, where LLM context parallelism is needed for the fused multimodal sequence but encoder inputs remain bounded. We pr"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.27678","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/2605.27678/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":"2605.27678","created_at":"2026-05-28T01:04:45.991663+00:00"},{"alias_kind":"arxiv_version","alias_value":"2605.27678v1","created_at":"2026-05-28T01:04:45.991663+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.27678","created_at":"2026-05-28T01:04:45.991663+00:00"},{"alias_kind":"pith_short_12","alias_value":"JSC4J7ABFLU4","created_at":"2026-05-28T01:04:45.991663+00:00"},{"alias_kind":"pith_short_16","alias_value":"JSC4J7ABFLU4NBQS","created_at":"2026-05-28T01:04:45.991663+00:00"},{"alias_kind":"pith_short_8","alias_value":"JSC4J7AB","created_at":"2026-05-28T01:04:45.991663+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/JSC4J7ABFLU4NBQSAHNZZZZPTT","json":"https://pith.science/pith/JSC4J7ABFLU4NBQSAHNZZZZPTT.json","graph_json":"https://pith.science/api/pith-number/JSC4J7ABFLU4NBQSAHNZZZZPTT/graph.json","events_json":"https://pith.science/api/pith-number/JSC4J7ABFLU4NBQSAHNZZZZPTT/events.json","paper":"https://pith.science/paper/JSC4J7AB"},"agent_actions":{"view_html":"https://pith.science/pith/JSC4J7ABFLU4NBQSAHNZZZZPTT","download_json":"https://pith.science/pith/JSC4J7ABFLU4NBQSAHNZZZZPTT.json","view_paper":"https://pith.science/paper/JSC4J7AB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2605.27678&json=true","fetch_graph":"https://pith.science/api/pith-number/JSC4J7ABFLU4NBQSAHNZZZZPTT/graph.json","fetch_events":"https://pith.science/api/pith-number/JSC4J7ABFLU4NBQSAHNZZZZPTT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/JSC4J7ABFLU4NBQSAHNZZZZPTT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/JSC4J7ABFLU4NBQSAHNZZZZPTT/action/storage_attestation","attest_author":"https://pith.science/pith/JSC4J7ABFLU4NBQSAHNZZZZPTT/action/author_attestation","sign_citation":"https://pith.science/pith/JSC4J7ABFLU4NBQSAHNZZZZPTT/action/citation_signature","submit_replication":"https://pith.science/pith/JSC4J7ABFLU4NBQSAHNZZZZPTT/action/replication_record"}},"created_at":"2026-05-28T01:04:45.991663+00:00","updated_at":"2026-05-28T01:04:45.991663+00:00"}