{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:3EZGRNNPVLYO4VQKWTI47E3ZWT","short_pith_number":"pith:3EZGRNNP","schema_version":"1.0","canonical_sha256":"d93268b5afaaf0ee560ab4d1cf9379b4c453610e115746d6aa299c0537bdc004","source":{"kind":"arxiv","id":"2606.05484","version":1},"attestation_state":"computed","paper":{"title":"Learned Subspace Compression for Communication-Efficient Pipeline Parallelism","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Edouard Oyallon, Eugene Belilovsky, Paul Janson","submitted_at":"2026-06-03T22:10:16Z","abstract_excerpt":"Pipeline parallelism enables training of large language models that exceed single-device memory, yet inter-stage activation communication becomes the dominant bottleneck when trained on low-bandwidth networks. Recent work in this area has proposed using fixed orthogonal projections to compress activations. However, this still results in a significant performance degradation and requires a number of non-standard adaptations to constrain the optimization. A natural alternative is to learn a low rank projection for each pipeline stage, however maintaining the necessary orthogonality of these proj"},"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":"2606.05484","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-06-03T22:10:16Z","cross_cats_sorted":[],"title_canon_sha256":"d4412d9bf2c4ce8a4514d0b0baf3adb3908f49484cd50f5d35f64eedb4d79ba2","abstract_canon_sha256":"b74d9e317de5d8a556ef77e6634555f6e95638ce1d0d96bd30e6c46906cf5748"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-05T01:14:52.618070Z","signature_b64":"bKFHwVODzFzEbti++OkruKHcUMZQcUnen1DMTQ86S1pxHFeALldfsj7IMa8i3U4ibsCG7ZBokl6i6MSLiysOBA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d93268b5afaaf0ee560ab4d1cf9379b4c453610e115746d6aa299c0537bdc004","last_reissued_at":"2026-06-05T01:14:52.617627Z","signature_status":"signed_v1","first_computed_at":"2026-06-05T01:14:52.617627Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Learned Subspace Compression for Communication-Efficient Pipeline Parallelism","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Edouard Oyallon, Eugene Belilovsky, Paul Janson","submitted_at":"2026-06-03T22:10:16Z","abstract_excerpt":"Pipeline parallelism enables training of large language models that exceed single-device memory, yet inter-stage activation communication becomes the dominant bottleneck when trained on low-bandwidth networks. Recent work in this area has proposed using fixed orthogonal projections to compress activations. However, this still results in a significant performance degradation and requires a number of non-standard adaptations to constrain the optimization. A natural alternative is to learn a low rank projection for each pipeline stage, however maintaining the necessary orthogonality of these proj"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.05484","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/2606.05484/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":"2606.05484","created_at":"2026-06-05T01:14:52.617689+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.05484v1","created_at":"2026-06-05T01:14:52.617689+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.05484","created_at":"2026-06-05T01:14:52.617689+00:00"},{"alias_kind":"pith_short_12","alias_value":"3EZGRNNPVLYO","created_at":"2026-06-05T01:14:52.617689+00:00"},{"alias_kind":"pith_short_16","alias_value":"3EZGRNNPVLYO4VQK","created_at":"2026-06-05T01:14:52.617689+00:00"},{"alias_kind":"pith_short_8","alias_value":"3EZGRNNP","created_at":"2026-06-05T01:14:52.617689+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/3EZGRNNPVLYO4VQKWTI47E3ZWT","json":"https://pith.science/pith/3EZGRNNPVLYO4VQKWTI47E3ZWT.json","graph_json":"https://pith.science/api/pith-number/3EZGRNNPVLYO4VQKWTI47E3ZWT/graph.json","events_json":"https://pith.science/api/pith-number/3EZGRNNPVLYO4VQKWTI47E3ZWT/events.json","paper":"https://pith.science/paper/3EZGRNNP"},"agent_actions":{"view_html":"https://pith.science/pith/3EZGRNNPVLYO4VQKWTI47E3ZWT","download_json":"https://pith.science/pith/3EZGRNNPVLYO4VQKWTI47E3ZWT.json","view_paper":"https://pith.science/paper/3EZGRNNP","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.05484&json=true","fetch_graph":"https://pith.science/api/pith-number/3EZGRNNPVLYO4VQKWTI47E3ZWT/graph.json","fetch_events":"https://pith.science/api/pith-number/3EZGRNNPVLYO4VQKWTI47E3ZWT/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3EZGRNNPVLYO4VQKWTI47E3ZWT/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3EZGRNNPVLYO4VQKWTI47E3ZWT/action/storage_attestation","attest_author":"https://pith.science/pith/3EZGRNNPVLYO4VQKWTI47E3ZWT/action/author_attestation","sign_citation":"https://pith.science/pith/3EZGRNNPVLYO4VQKWTI47E3ZWT/action/citation_signature","submit_replication":"https://pith.science/pith/3EZGRNNPVLYO4VQKWTI47E3ZWT/action/replication_record"}},"created_at":"2026-06-05T01:14:52.617689+00:00","updated_at":"2026-06-05T01:14:52.617689+00:00"}