{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:USNR7PI7ZDEQNRM6DFJQ3IZ4SX","short_pith_number":"pith:USNR7PI7","canonical_record":{"source":{"id":"2605.18404","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.DC","submitted_at":"2026-05-18T13:45:58Z","cross_cats_sorted":[],"title_canon_sha256":"5804b1dc7c8ddcdc6deccf583a7fba00c7c12204df1953ab2a3e4393096b31d6","abstract_canon_sha256":"9fecb13af08e5670227cceed92709b3fb86475c7ed367aa5af0c2b5111645675"},"schema_version":"1.0"},"canonical_sha256":"a49b1fbd1fc8c906c59e19530da33c95cf5a0e31e6cdf40bfce79846c10792f1","source":{"kind":"arxiv","id":"2605.18404","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.18404","created_at":"2026-05-20T00:05:59Z"},{"alias_kind":"arxiv_version","alias_value":"2605.18404v1","created_at":"2026-05-20T00:05:59Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.18404","created_at":"2026-05-20T00:05:59Z"},{"alias_kind":"pith_short_12","alias_value":"USNR7PI7ZDEQ","created_at":"2026-05-20T00:05:59Z"},{"alias_kind":"pith_short_16","alias_value":"USNR7PI7ZDEQNRM6","created_at":"2026-05-20T00:05:59Z"},{"alias_kind":"pith_short_8","alias_value":"USNR7PI7","created_at":"2026-05-20T00:05:59Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:USNR7PI7ZDEQNRM6DFJQ3IZ4SX","target":"record","payload":{"canonical_record":{"source":{"id":"2605.18404","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.DC","submitted_at":"2026-05-18T13:45:58Z","cross_cats_sorted":[],"title_canon_sha256":"5804b1dc7c8ddcdc6deccf583a7fba00c7c12204df1953ab2a3e4393096b31d6","abstract_canon_sha256":"9fecb13af08e5670227cceed92709b3fb86475c7ed367aa5af0c2b5111645675"},"schema_version":"1.0"},"canonical_sha256":"a49b1fbd1fc8c906c59e19530da33c95cf5a0e31e6cdf40bfce79846c10792f1","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T00:05:59.052579Z","signature_b64":"3Q6lnAFCJv63/cCvC+pC1UpRDdIu0ZOOv/bY/D7c9cVXW0DZDeu6+StpStQAk0Gh/ZjR9qqFCNrmDLkpvpT/CA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a49b1fbd1fc8c906c59e19530da33c95cf5a0e31e6cdf40bfce79846c10792f1","last_reissued_at":"2026-05-20T00:05:59.051865Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T00:05:59.051865Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.18404","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-20T00:05:59Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Vu8P3OxbHLAtd2Z17aTa6Tn1HMPeZ2kcM6czY584Rxnzp1qgtNvce6JoWl+9OL1lWUE3tFmu4LUAeLh2YQHaAA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-24T06:12:38.137033Z"},"content_sha256":"6758d37cd18936cd81f66eaea6d824a425f7bfa4efc2de625f017d55a5f2e7fb","schema_version":"1.0","event_id":"sha256:6758d37cd18936cd81f66eaea6d824a425f7bfa4efc2de625f017d55a5f2e7fb"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:USNR7PI7ZDEQNRM6DFJQ3IZ4SX","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"JanusPipe: Efficient Pipeline Parallel Training for Machine Learning Interatomic Potentials","license":"http://creativecommons.org/licenses/by/4.0/","headline":"JanusPipe enables efficient pipeline parallel training for conservative MLIPs by using SymFold and WaveK to handle their double-backward pattern.","cross_cats":[],"primary_cat":"cs.DC","authors_text":"Guangming Tan, Hongtao Xu, Hongyu Wang, Mingzhen Li, Weijian Liu, Weile Jia, Yan Wang","submitted_at":"2026-05-18T13:45:58Z","abstract_excerpt":"Discovering atom-level phenomena requires molecular dynamics (MD) simulations with ab initio accuracy. Machine learning interatomic potentials (MLIPs) enable stable, high-accuracy MD simulations, and their models exhibit scaling-law trends similar to large language models. However, the lack of scalable and efficient distributed training systems for conservative MLIPs makes them difficult to scale. This is because conservative MLIPs inherently follow a double-backward execution pattern, which involves computing gradients during the forward pass. This pattern creates a mismatch with existing dis"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We present JanusPipe, an efficient 3D-parallel (PP/DP/GP) training system tailored for conservative MLIPs. It integrates SymFold to enable memory-efficient pipeline parallelism for conservative MLIPs, and WaveK to reduce pipeline bubbles by balancing the four-phase compute time. Experimental results on 32 GPUs show that JanusPipe improves throughput by 1.51× and 1.45× on average over 1F1B and Hanayo, respectively.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The double-backward execution pattern of conservative MLIPs creates a fundamental mismatch with existing pipeline parallelism systems that SymFold and WaveK can resolve without introducing significant accuracy loss, overhead, or model-specific constraints that limit generality.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"JanusPipe is a new 3D-parallel training system for conservative MLIPs that uses SymFold and WaveK to achieve 1.51x and 1.45x average throughput gains over 1F1B and Hanayo on 32 GPUs.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"JanusPipe enables efficient pipeline parallel training for conservative MLIPs by using SymFold and WaveK to handle their double-backward pattern.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"5f58c3cb6f46073f2b874991782a27217821c48c208ede8d33c8e84e7c873b40"},"source":{"id":"2605.18404","kind":"arxiv","version":1},"verdict":{"id":"973cebf9-e093-4536-b9e0-3fec11579464","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-19T23:45:29.081282Z","strongest_claim":"We present JanusPipe, an efficient 3D-parallel (PP/DP/GP) training system tailored for conservative MLIPs. It integrates SymFold to enable memory-efficient pipeline parallelism for conservative MLIPs, and WaveK to reduce pipeline bubbles by balancing the four-phase compute time. Experimental results on 32 GPUs show that JanusPipe improves throughput by 1.51× and 1.45× on average over 1F1B and Hanayo, respectively.","one_line_summary":"JanusPipe is a new 3D-parallel training system for conservative MLIPs that uses SymFold and WaveK to achieve 1.51x and 1.45x average throughput gains over 1F1B and Hanayo on 32 GPUs.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The double-backward execution pattern of conservative MLIPs creates a fundamental mismatch with existing pipeline parallelism systems that SymFold and WaveK can resolve without introducing significant accuracy loss, overhead, or model-specific constraints that limit generality.","pith_extraction_headline":"JanusPipe enables efficient pipeline parallel training for conservative MLIPs by using SymFold and WaveK to handle their double-backward pattern."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.18404/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_title_agreement","ran_at":"2026-05-20T00:01:20.325798Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"cited_work_retraction","ran_at":"2026-05-19T23:52:09.926291Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T23:51:43.418457Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"citation_quote_validity","ran_at":"2026-05-19T23:50:02.474945Z","status":"skipped","version":"0.1.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T23:33:29.268841Z","status":"skipped","version":"1.0.0","findings_count":0},{"name":"external_links","ran_at":"2026-05-19T23:31:39.789069Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"claim_evidence","ran_at":"2026-05-19T23:21:58.715676Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"ac6f27ba98229ded44e4df8b233e8226bee46face4da651115c17d5d39b2090e"},"references":{"count":94,"sample":[{"doi":"10.48550/arxiv.2401.00096","year":null,"title":"A foundation model for atomistic materials chemistry","work_id":"29bfbd6f-762e-4448-8f07-4f24bf553e51","ref_index":1,"cited_arxiv_id":"2401.00096","is_internal_anchor":true},{"doi":"10.48550/arxiv.2410.12771","year":2024,"title":"Open Materials 2024 (OMat24) Inorganic Materials Dataset and Models","work_id":"22935efd-40fe-48d5-974f-f4c033759df0","ref_index":2,"cited_arxiv_id":"2410.12771","is_internal_anchor":true},{"doi":"10.48550/arxiv.2405.04967","year":null,"title":"MatterSim: A Deep Learning Atomistic Model Across Elements, Temperatures and Pressures","work_id":"ee6227fe-2fc7-49bc-990c-aae7db05c4a4","ref_index":3,"cited_arxiv_id":"2405.04967","is_internal_anchor":true},{"doi":"","year":null,"title":"Forty-second International Conference on Machine Learning , year=","work_id":"c0b99e90-4b7a-486c-838d-58f42e16ec74","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2010,"title":"Tuckerman, Mark E. , title =. 2010 , address =","work_id":"2a0fc52a-9573-4bc5-af15-8139b29d1430","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":94,"snapshot_sha256":"e93bab2027374c4dabaea6a2ba2287dbe0252a34c68d7a4253c511a09aa72513","internal_anchors":10},"formal_canon":{"evidence_count":2,"snapshot_sha256":"e25091c7e2a1e274865f27bab646304edb65ff0efcc0697f29b0396a7ee1c3bd"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"973cebf9-e093-4536-b9e0-3fec11579464"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-20T00:05:59Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"JtfSxYsahUVYwrn5IVKAunWnxWCt0qOdgAVYdz1xNdeFGuwy5OJiQGtR+jhdJmHbdoh+iZpr3aRePmxsRUsFCw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-24T06:12:38.138334Z"},"content_sha256":"c554d0d5ad76023e2f32bdd5c06b369f8ac40e845ffeb3ca7606c4b6b3376faa","schema_version":"1.0","event_id":"sha256:c554d0d5ad76023e2f32bdd5c06b369f8ac40e845ffeb3ca7606c4b6b3376faa"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/USNR7PI7ZDEQNRM6DFJQ3IZ4SX/bundle.json","state_url":"https://pith.science/pith/USNR7PI7ZDEQNRM6DFJQ3IZ4SX/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/USNR7PI7ZDEQNRM6DFJQ3IZ4SX/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":"2026-05-24T06:12:38Z","links":{"resolver":"https://pith.science/pith/USNR7PI7ZDEQNRM6DFJQ3IZ4SX","bundle":"https://pith.science/pith/USNR7PI7ZDEQNRM6DFJQ3IZ4SX/bundle.json","state":"https://pith.science/pith/USNR7PI7ZDEQNRM6DFJQ3IZ4SX/state.json","well_known_bundle":"https://pith.science/.well-known/pith/USNR7PI7ZDEQNRM6DFJQ3IZ4SX/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:USNR7PI7ZDEQNRM6DFJQ3IZ4SX","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":"9fecb13af08e5670227cceed92709b3fb86475c7ed367aa5af0c2b5111645675","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.DC","submitted_at":"2026-05-18T13:45:58Z","title_canon_sha256":"5804b1dc7c8ddcdc6deccf583a7fba00c7c12204df1953ab2a3e4393096b31d6"},"schema_version":"1.0","source":{"id":"2605.18404","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.18404","created_at":"2026-05-20T00:05:59Z"},{"alias_kind":"arxiv_version","alias_value":"2605.18404v1","created_at":"2026-05-20T00:05:59Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.18404","created_at":"2026-05-20T00:05:59Z"},{"alias_kind":"pith_short_12","alias_value":"USNR7PI7ZDEQ","created_at":"2026-05-20T00:05:59Z"},{"alias_kind":"pith_short_16","alias_value":"USNR7PI7ZDEQNRM6","created_at":"2026-05-20T00:05:59Z"},{"alias_kind":"pith_short_8","alias_value":"USNR7PI7","created_at":"2026-05-20T00:05:59Z"}],"graph_snapshots":[{"event_id":"sha256:c554d0d5ad76023e2f32bdd5c06b369f8ac40e845ffeb3ca7606c4b6b3376faa","target":"graph","created_at":"2026-05-20T00:05:59Z","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":"We present JanusPipe, an efficient 3D-parallel (PP/DP/GP) training system tailored for conservative MLIPs. It integrates SymFold to enable memory-efficient pipeline parallelism for conservative MLIPs, and WaveK to reduce pipeline bubbles by balancing the four-phase compute time. Experimental results on 32 GPUs show that JanusPipe improves throughput by 1.51× and 1.45× on average over 1F1B and Hanayo, respectively."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The double-backward execution pattern of conservative MLIPs creates a fundamental mismatch with existing pipeline parallelism systems that SymFold and WaveK can resolve without introducing significant accuracy loss, overhead, or model-specific constraints that limit generality."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"JanusPipe is a new 3D-parallel training system for conservative MLIPs that uses SymFold and WaveK to achieve 1.51x and 1.45x average throughput gains over 1F1B and Hanayo on 32 GPUs."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"JanusPipe enables efficient pipeline parallel training for conservative MLIPs by using SymFold and WaveK to handle their double-backward pattern."}],"snapshot_sha256":"5f58c3cb6f46073f2b874991782a27217821c48c208ede8d33c8e84e7c873b40"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"e25091c7e2a1e274865f27bab646304edb65ff0efcc0697f29b0396a7ee1c3bd"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"doi_title_agreement","ran_at":"2026-05-20T00:01:20.325798Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"cited_work_retraction","ran_at":"2026-05-19T23:52:09.926291Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-19T23:51:43.418457Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"citation_quote_validity","ran_at":"2026-05-19T23:50:02.474945Z","status":"skipped","version":"0.1.0"},{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-19T23:33:29.268841Z","status":"skipped","version":"1.0.0"},{"findings_count":0,"name":"external_links","ran_at":"2026-05-19T23:31:39.789069Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"claim_evidence","ran_at":"2026-05-19T23:21:58.715676Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2605.18404/integrity.json","findings":[],"snapshot_sha256":"ac6f27ba98229ded44e4df8b233e8226bee46face4da651115c17d5d39b2090e","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Discovering atom-level phenomena requires molecular dynamics (MD) simulations with ab initio accuracy. Machine learning interatomic potentials (MLIPs) enable stable, high-accuracy MD simulations, and their models exhibit scaling-law trends similar to large language models. However, the lack of scalable and efficient distributed training systems for conservative MLIPs makes them difficult to scale. This is because conservative MLIPs inherently follow a double-backward execution pattern, which involves computing gradients during the forward pass. This pattern creates a mismatch with existing dis","authors_text":"Guangming Tan, Hongtao Xu, Hongyu Wang, Mingzhen Li, Weijian Liu, Weile Jia, Yan Wang","cross_cats":[],"headline":"JanusPipe enables efficient pipeline parallel training for conservative MLIPs by using SymFold and WaveK to handle their double-backward pattern.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.DC","submitted_at":"2026-05-18T13:45:58Z","title":"JanusPipe: Efficient Pipeline Parallel Training for Machine Learning Interatomic Potentials"},"references":{"count":94,"internal_anchors":10,"resolved_work":94,"sample":[{"cited_arxiv_id":"2401.00096","doi":"10.48550/arxiv.2401.00096","is_internal_anchor":true,"ref_index":1,"title":"A foundation model for atomistic materials chemistry","work_id":"29bfbd6f-762e-4448-8f07-4f24bf553e51","year":null},{"cited_arxiv_id":"2410.12771","doi":"10.48550/arxiv.2410.12771","is_internal_anchor":true,"ref_index":2,"title":"Open Materials 2024 (OMat24) Inorganic Materials Dataset and Models","work_id":"22935efd-40fe-48d5-974f-f4c033759df0","year":2024},{"cited_arxiv_id":"2405.04967","doi":"10.48550/arxiv.2405.04967","is_internal_anchor":true,"ref_index":3,"title":"MatterSim: A Deep Learning Atomistic Model Across Elements, Temperatures and Pressures","work_id":"ee6227fe-2fc7-49bc-990c-aae7db05c4a4","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Forty-second International Conference on Machine Learning , year=","work_id":"c0b99e90-4b7a-486c-838d-58f42e16ec74","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Tuckerman, Mark E. , title =. 2010 , address =","work_id":"2a0fc52a-9573-4bc5-af15-8139b29d1430","year":2010}],"snapshot_sha256":"e93bab2027374c4dabaea6a2ba2287dbe0252a34c68d7a4253c511a09aa72513"},"source":{"id":"2605.18404","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-19T23:45:29.081282Z","id":"973cebf9-e093-4536-b9e0-3fec11579464","model_set":{"reader":"grok-4.3"},"one_line_summary":"JanusPipe is a new 3D-parallel training system for conservative MLIPs that uses SymFold and WaveK to achieve 1.51x and 1.45x average throughput gains over 1F1B and Hanayo on 32 GPUs.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"JanusPipe enables efficient pipeline parallel training for conservative MLIPs by using SymFold and WaveK to handle their double-backward pattern.","strongest_claim":"We present JanusPipe, an efficient 3D-parallel (PP/DP/GP) training system tailored for conservative MLIPs. It integrates SymFold to enable memory-efficient pipeline parallelism for conservative MLIPs, and WaveK to reduce pipeline bubbles by balancing the four-phase compute time. Experimental results on 32 GPUs show that JanusPipe improves throughput by 1.51× and 1.45× on average over 1F1B and Hanayo, respectively.","weakest_assumption":"The double-backward execution pattern of conservative MLIPs creates a fundamental mismatch with existing pipeline parallelism systems that SymFold and WaveK can resolve without introducing significant accuracy loss, overhead, or model-specific constraints that limit generality."}},"verdict_id":"973cebf9-e093-4536-b9e0-3fec11579464"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:6758d37cd18936cd81f66eaea6d824a425f7bfa4efc2de625f017d55a5f2e7fb","target":"record","created_at":"2026-05-20T00:05:59Z","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":"9fecb13af08e5670227cceed92709b3fb86475c7ed367aa5af0c2b5111645675","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.DC","submitted_at":"2026-05-18T13:45:58Z","title_canon_sha256":"5804b1dc7c8ddcdc6deccf583a7fba00c7c12204df1953ab2a3e4393096b31d6"},"schema_version":"1.0","source":{"id":"2605.18404","kind":"arxiv","version":1}},"canonical_sha256":"a49b1fbd1fc8c906c59e19530da33c95cf5a0e31e6cdf40bfce79846c10792f1","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"a49b1fbd1fc8c906c59e19530da33c95cf5a0e31e6cdf40bfce79846c10792f1","first_computed_at":"2026-05-20T00:05:59.051865Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-20T00:05:59.051865Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"3Q6lnAFCJv63/cCvC+pC1UpRDdIu0ZOOv/bY/D7c9cVXW0DZDeu6+StpStQAk0Gh/ZjR9qqFCNrmDLkpvpT/CA==","signature_status":"signed_v1","signed_at":"2026-05-20T00:05:59.052579Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.18404","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:6758d37cd18936cd81f66eaea6d824a425f7bfa4efc2de625f017d55a5f2e7fb","sha256:c554d0d5ad76023e2f32bdd5c06b369f8ac40e845ffeb3ca7606c4b6b3376faa"],"state_sha256":"b4331041c285795226066d0eed575f1a3bb98a7510d3b67426a18c93ef3dff5b"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"xFD9BvmJGQci9TvmsbTsZ7SZjpTHmEf/piHNDeaatNKGF3zF0iJc88HJo6T6Wi29js2ys++9l3yjAwTBAg9kAg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-24T06:12:38.143363Z","bundle_sha256":"757b3a921288e9c7defd349b7c975e44c839d6c901b87f1ef4bbb73a48e06c1b"}}