{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:22A2GXQDAIZK6PQBG5ITGIDVYI","short_pith_number":"pith:22A2GXQD","canonical_record":{"source":{"id":"2602.03839","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-02-03T18:56:48Z","cross_cats_sorted":[],"title_canon_sha256":"4c4d8c2ab8108fc12ec7a228cd954361d99e1a9a2ffa7930cf62ddc7f5e1b0a9","abstract_canon_sha256":"7c16bad798a4a1c8645bdeaaca7e7d37f402d7210c093e114ba2cabdfa12235c"},"schema_version":"1.0"},"canonical_sha256":"d681a35e030232af3e013751332075c20a58e3ea480e2af4d2858c02ab1baabe","source":{"kind":"arxiv","id":"2602.03839","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2602.03839","created_at":"2026-05-20T02:05:40Z"},{"alias_kind":"arxiv_version","alias_value":"2602.03839v2","created_at":"2026-05-20T02:05:40Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.03839","created_at":"2026-05-20T02:05:40Z"},{"alias_kind":"pith_short_12","alias_value":"22A2GXQDAIZK","created_at":"2026-05-20T02:05:40Z"},{"alias_kind":"pith_short_16","alias_value":"22A2GXQDAIZK6PQB","created_at":"2026-05-20T02:05:40Z"},{"alias_kind":"pith_short_8","alias_value":"22A2GXQD","created_at":"2026-05-20T02:05:40Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:22A2GXQDAIZK6PQBG5ITGIDVYI","target":"record","payload":{"canonical_record":{"source":{"id":"2602.03839","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-02-03T18:56:48Z","cross_cats_sorted":[],"title_canon_sha256":"4c4d8c2ab8108fc12ec7a228cd954361d99e1a9a2ffa7930cf62ddc7f5e1b0a9","abstract_canon_sha256":"7c16bad798a4a1c8645bdeaaca7e7d37f402d7210c093e114ba2cabdfa12235c"},"schema_version":"1.0"},"canonical_sha256":"d681a35e030232af3e013751332075c20a58e3ea480e2af4d2858c02ab1baabe","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-20T02:05:40.248898Z","signature_b64":"V+bkz9N4/a/H1nkgV3IaM//PhKu0UkrGv5wQmVEwDCKUmM9XcToeUPjZO8zpwMGrv1cd56nkIC8R5oDXYGGKDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d681a35e030232af3e013751332075c20a58e3ea480e2af4d2858c02ab1baabe","last_reissued_at":"2026-05-20T02:05:40.248082Z","signature_status":"signed_v1","first_computed_at":"2026-05-20T02:05:40.248082Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2602.03839","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-20T02:05:40Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"wM0yYGvryNMSTd8fitZxQfq7cnY4qR0YBQxcqk5LkaKSSSJMayDOfvdP+lmP9I3s6/oD3yJ/c8TTBqAX1xVuDA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-22T16:54:54.195027Z"},"content_sha256":"c2e8f79f3fc2c967677e590518e4153e6b12adf6020581fcb570c0d0c5ba9a13","schema_version":"1.0","event_id":"sha256:c2e8f79f3fc2c967677e590518e4153e6b12adf6020581fcb570c0d0c5ba9a13"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:22A2GXQDAIZK6PQBG5ITGIDVYI","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Understanding and Exploiting Weight Update Sparsity for Communication-Efficient Distributed RL","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Erfan Miahi, Eugene Belilovsky","submitted_at":"2026-02-03T18:56:48Z","abstract_excerpt":"Bandwidth-constrained distributed reinforcement learning (RL) post-training of large language models is bottlenecked by two channels: weight synchronization from trainers to inference workers, and gradient or pseudo-gradient synchronization across trainers. We find that approximately 99% of per-step weight updates are invisible after the BF16 cast used by standard training and inference forward passes. We explain this sparsity by showing that, at typical RL post-training learning rates, Adam updates often fall below the local BF16 rounding threshold. We turn this observation into an algorithmi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.03839","kind":"arxiv","version":2},"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/2602.03839/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"},"verdict_id":null},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-20T02:05:40Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"SfsptUXMf3Ss820e6+/Cf3d3MAik+LOetJBb38LrAFmdQJ86zm7usy8icnn69TxAkprsQx1FfnEN6nPoipixCQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-22T16:54:54.195738Z"},"content_sha256":"b6fbd03663f983e628fabfa28e37130fb8c342e63fb3ceaa0739bfc72a0e3f99","schema_version":"1.0","event_id":"sha256:b6fbd03663f983e628fabfa28e37130fb8c342e63fb3ceaa0739bfc72a0e3f99"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/22A2GXQDAIZK6PQBG5ITGIDVYI/bundle.json","state_url":"https://pith.science/pith/22A2GXQDAIZK6PQBG5ITGIDVYI/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/22A2GXQDAIZK6PQBG5ITGIDVYI/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":"2026-05-22T16:54:54Z","links":{"resolver":"https://pith.science/pith/22A2GXQDAIZK6PQBG5ITGIDVYI","bundle":"https://pith.science/pith/22A2GXQDAIZK6PQBG5ITGIDVYI/bundle.json","state":"https://pith.science/pith/22A2GXQDAIZK6PQBG5ITGIDVYI/state.json","well_known_bundle":"https://pith.science/.well-known/pith/22A2GXQDAIZK6PQBG5ITGIDVYI/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:22A2GXQDAIZK6PQBG5ITGIDVYI","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":"7c16bad798a4a1c8645bdeaaca7e7d37f402d7210c093e114ba2cabdfa12235c","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-02-03T18:56:48Z","title_canon_sha256":"4c4d8c2ab8108fc12ec7a228cd954361d99e1a9a2ffa7930cf62ddc7f5e1b0a9"},"schema_version":"1.0","source":{"id":"2602.03839","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2602.03839","created_at":"2026-05-20T02:05:40Z"},{"alias_kind":"arxiv_version","alias_value":"2602.03839v2","created_at":"2026-05-20T02:05:40Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.03839","created_at":"2026-05-20T02:05:40Z"},{"alias_kind":"pith_short_12","alias_value":"22A2GXQDAIZK","created_at":"2026-05-20T02:05:40Z"},{"alias_kind":"pith_short_16","alias_value":"22A2GXQDAIZK6PQB","created_at":"2026-05-20T02:05:40Z"},{"alias_kind":"pith_short_8","alias_value":"22A2GXQD","created_at":"2026-05-20T02:05:40Z"}],"graph_snapshots":[{"event_id":"sha256:b6fbd03663f983e628fabfa28e37130fb8c342e63fb3ceaa0739bfc72a0e3f99","target":"graph","created_at":"2026-05-20T02:05:40Z","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":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2602.03839/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Bandwidth-constrained distributed reinforcement learning (RL) post-training of large language models is bottlenecked by two channels: weight synchronization from trainers to inference workers, and gradient or pseudo-gradient synchronization across trainers. We find that approximately 99% of per-step weight updates are invisible after the BF16 cast used by standard training and inference forward passes. We explain this sparsity by showing that, at typical RL post-training learning rates, Adam updates often fall below the local BF16 rounding threshold. We turn this observation into an algorithmi","authors_text":"Erfan Miahi, Eugene Belilovsky","cross_cats":[],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-02-03T18:56:48Z","title":"Understanding and Exploiting Weight Update Sparsity for Communication-Efficient Distributed RL"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.03839","kind":"arxiv","version":2},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:c2e8f79f3fc2c967677e590518e4153e6b12adf6020581fcb570c0d0c5ba9a13","target":"record","created_at":"2026-05-20T02:05:40Z","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":"7c16bad798a4a1c8645bdeaaca7e7d37f402d7210c093e114ba2cabdfa12235c","cross_cats_sorted":[],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-02-03T18:56:48Z","title_canon_sha256":"4c4d8c2ab8108fc12ec7a228cd954361d99e1a9a2ffa7930cf62ddc7f5e1b0a9"},"schema_version":"1.0","source":{"id":"2602.03839","kind":"arxiv","version":2}},"canonical_sha256":"d681a35e030232af3e013751332075c20a58e3ea480e2af4d2858c02ab1baabe","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"d681a35e030232af3e013751332075c20a58e3ea480e2af4d2858c02ab1baabe","first_computed_at":"2026-05-20T02:05:40.248082Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-20T02:05:40.248082Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"V+bkz9N4/a/H1nkgV3IaM//PhKu0UkrGv5wQmVEwDCKUmM9XcToeUPjZO8zpwMGrv1cd56nkIC8R5oDXYGGKDQ==","signature_status":"signed_v1","signed_at":"2026-05-20T02:05:40.248898Z","signed_message":"canonical_sha256_bytes"},"source_id":"2602.03839","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:c2e8f79f3fc2c967677e590518e4153e6b12adf6020581fcb570c0d0c5ba9a13","sha256:b6fbd03663f983e628fabfa28e37130fb8c342e63fb3ceaa0739bfc72a0e3f99"],"state_sha256":"b7fe76430ea3ad1d248a3ff00ec4bf7fb77c6934d2274c3b0371abc17dd6e8cb"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"XVhHE2S1dLs+kQGIyjbulJG79IAY0onm/ftyN90njofvPYkgoLsLhf3QVOsz2Xh7gqR3ngH7n99bpwlkZCzHCg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-22T16:54:54.199258Z","bundle_sha256":"a8beef33be04af33397a8105ff8e73cf4a0af70967ef2cb425d0c7810e0efb78"}}