{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:DW7ZUEYWYDX4DDTFZSFGPPPXPM","short_pith_number":"pith:DW7ZUEYW","schema_version":"1.0","canonical_sha256":"1dbf9a1316c0efc18e65cc8a67bdf77b3595f9f171bca1b4bb92dc4881eeb2c9","source":{"kind":"arxiv","id":"2508.15706","version":3},"attestation_state":"computed","paper":{"title":"Overcoming the Communication-Performance Tradeoff in LLM Pretraining","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Amir Sarfi, Benjamin Th\\'erien, Eugene Belilovsky, Joel Lidin","submitted_at":"2025-08-21T16:48:19Z","abstract_excerpt":"Communication-efficient distributed training algorithms (e.g., DiLoCo) have received considerable interest due to their benefits for training large language models (LLMs) in bandwidth-constrained settings, such as across datacenters and over the internet. While these local-update methods achieve communication reduction through reduced synchronization, they still require communicating dense model-sized pseudo-gradients, resulting in a communication bottleneck even for cross-datacenter links. While quantization is often applied to reduce the pseudo-gradient's size, in the context of LLM pre-trai"},"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":"2508.15706","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-08-21T16:48:19Z","cross_cats_sorted":[],"title_canon_sha256":"ccf652e6459adc3ce58bfc02a64497cb64964de4fc20e3d27ea2dee69d082f8a","abstract_canon_sha256":"4f67c3079c4fd8251f5ab6b604b36ae11df9726e0a315a08e6aa4c50dcb78baf"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-24T01:22:59.673096Z","signature_b64":"sqj1r50FEQysoj+LBIXrdF51W7/ohXEr2T1LZusNMgPkfNfMwtLyDwzAmlM4+jPGVoA8jWRZg9LePgDdx37PCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1dbf9a1316c0efc18e65cc8a67bdf77b3595f9f171bca1b4bb92dc4881eeb2c9","last_reissued_at":"2026-07-24T01:22:59.672093Z","signature_status":"signed_v1","first_computed_at":"2026-07-24T01:22:59.672093Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Overcoming the Communication-Performance Tradeoff in LLM Pretraining","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Amir Sarfi, Benjamin Th\\'erien, Eugene Belilovsky, Joel Lidin","submitted_at":"2025-08-21T16:48:19Z","abstract_excerpt":"Communication-efficient distributed training algorithms (e.g., DiLoCo) have received considerable interest due to their benefits for training large language models (LLMs) in bandwidth-constrained settings, such as across datacenters and over the internet. While these local-update methods achieve communication reduction through reduced synchronization, they still require communicating dense model-sized pseudo-gradients, resulting in a communication bottleneck even for cross-datacenter links. While quantization is often applied to reduce the pseudo-gradient's size, in the context of LLM pre-trai"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.15706","kind":"arxiv","version":3},"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/2508.15706/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":"2508.15706","created_at":"2026-07-24T01:22:59.672545+00:00"},{"alias_kind":"arxiv_version","alias_value":"2508.15706v3","created_at":"2026-07-24T01:22:59.672545+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.15706","created_at":"2026-07-24T01:22:59.672545+00:00"},{"alias_kind":"pith_short_12","alias_value":"DW7ZUEYWYDX4","created_at":"2026-07-24T01:22:59.672545+00:00"},{"alias_kind":"pith_short_16","alias_value":"DW7ZUEYWYDX4DDTF","created_at":"2026-07-24T01:22:59.672545+00:00"},{"alias_kind":"pith_short_8","alias_value":"DW7ZUEYW","created_at":"2026-07-24T01:22:59.672545+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":7,"internal_anchor_count":7,"sample":[{"citing_arxiv_id":"2606.11081","citing_title":"Unifying Local Communications and Local Updates for LLM Pretraining","ref_index":29,"is_internal_anchor":true},{"citing_arxiv_id":"2606.05484","citing_title":"Learned Subspace Compression for Communication-Efficient Pipeline Parallelism","ref_index":46,"is_internal_anchor":true},{"citing_arxiv_id":"2606.19025","citing_title":"FoMoE: Breaking the Full-Replica Barrier with a Federation of MoEs","ref_index":143,"is_internal_anchor":true},{"citing_arxiv_id":"2602.03839","citing_title":"Understanding and Exploiting Weight Update Sparsity for Communication-Efficient Distributed RL","ref_index":19,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06534","citing_title":"ROSE: Rollout On Serving GPUs via Cooperative Elasticity for Agentic RL","ref_index":54,"is_internal_anchor":true},{"citing_arxiv_id":"2605.06534","citing_title":"ROSE: Rollout On Serving GPUs via Cooperative Elasticity for Agentic RL","ref_index":55,"is_internal_anchor":true},{"citing_arxiv_id":"2604.11947","citing_title":"ResBM: Residual Bottleneck Models for Low-Bandwidth Pipeline Parallelism","ref_index":8,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/DW7ZUEYWYDX4DDTFZSFGPPPXPM","json":"https://pith.science/pith/DW7ZUEYWYDX4DDTFZSFGPPPXPM.json","graph_json":"https://pith.science/api/pith-number/DW7ZUEYWYDX4DDTFZSFGPPPXPM/graph.json","events_json":"https://pith.science/api/pith-number/DW7ZUEYWYDX4DDTFZSFGPPPXPM/events.json","paper":"https://pith.science/paper/DW7ZUEYW"},"agent_actions":{"view_html":"https://pith.science/pith/DW7ZUEYWYDX4DDTFZSFGPPPXPM","download_json":"https://pith.science/pith/DW7ZUEYWYDX4DDTFZSFGPPPXPM.json","view_paper":"https://pith.science/paper/DW7ZUEYW","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2508.15706&json=true","fetch_graph":"https://pith.science/api/pith-number/DW7ZUEYWYDX4DDTFZSFGPPPXPM/graph.json","fetch_events":"https://pith.science/api/pith-number/DW7ZUEYWYDX4DDTFZSFGPPPXPM/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DW7ZUEYWYDX4DDTFZSFGPPPXPM/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DW7ZUEYWYDX4DDTFZSFGPPPXPM/action/storage_attestation","attest_author":"https://pith.science/pith/DW7ZUEYWYDX4DDTFZSFGPPPXPM/action/author_attestation","sign_citation":"https://pith.science/pith/DW7ZUEYWYDX4DDTFZSFGPPPXPM/action/citation_signature","submit_replication":"https://pith.science/pith/DW7ZUEYWYDX4DDTFZSFGPPPXPM/action/replication_record"}},"created_at":"2026-07-24T01:22:59.672545+00:00","updated_at":"2026-07-24T01:22:59.672545+00:00"}