{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2024:NUGXK6Y4YSCE545ILLRI6N2ZGR","short_pith_number":"pith:NUGXK6Y4","canonical_record":{"source":{"id":"2409.12517","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2024-09-19T07:15:58Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"9a1a7ec3f7c6beb04465ea9b3a27caf24513566228e1ed75f298132dd59f7611","abstract_canon_sha256":"9d934df82c6b51ac1b5c36f22290ff649864a57d70fa5ced999bb575b0be72aa"},"schema_version":"1.0"},"canonical_sha256":"6d0d757b1cc4844ef3a85ae28f37593443d6dfa2b6d9217075eb06fc2e7b7c31","source":{"kind":"arxiv","id":"2409.12517","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2409.12517","created_at":"2026-07-05T10:11:24Z"},{"alias_kind":"arxiv_version","alias_value":"2409.12517v2","created_at":"2026-07-05T10:11:24Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2409.12517","created_at":"2026-07-05T10:11:24Z"},{"alias_kind":"pith_short_12","alias_value":"NUGXK6Y4YSCE","created_at":"2026-07-05T10:11:24Z"},{"alias_kind":"pith_short_16","alias_value":"NUGXK6Y4YSCE545I","created_at":"2026-07-05T10:11:24Z"},{"alias_kind":"pith_short_8","alias_value":"NUGXK6Y4","created_at":"2026-07-05T10:11:24Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2024:NUGXK6Y4YSCE545ILLRI6N2ZGR","target":"record","payload":{"canonical_record":{"source":{"id":"2409.12517","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2024-09-19T07:15:58Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"9a1a7ec3f7c6beb04465ea9b3a27caf24513566228e1ed75f298132dd59f7611","abstract_canon_sha256":"9d934df82c6b51ac1b5c36f22290ff649864a57d70fa5ced999bb575b0be72aa"},"schema_version":"1.0"},"canonical_sha256":"6d0d757b1cc4844ef3a85ae28f37593443d6dfa2b6d9217075eb06fc2e7b7c31","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:11:24.427284Z","signature_b64":"jrbWFqPR0aOEZc/wF1H/uG4hFNJxb1ia4+IptCU4jT5TTqkidd04qZKAxpE0SfmazvzEByF+1cc+uWRdIPx6Ag==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6d0d757b1cc4844ef3a85ae28f37593443d6dfa2b6d9217075eb06fc2e7b7c31","last_reissued_at":"2026-07-05T10:11:24.426732Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:11:24.426732Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2409.12517","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":"2026-07-05T10:11:24Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"dsshOM2kX+IiH79asYW2NHmQbF4fJB2MQZGwtwqSqpNZl89fntQSMrNQddYjfPS+3kCus0IcklRWg0PXYS7WAA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-03T10:58:17.206950Z"},"content_sha256":"ecd2aad0897666b20d01a97ed33e6fd41655e372098cddc8c9480b7cd8201ca4","schema_version":"1.0","event_id":"sha256:ecd2aad0897666b20d01a97ed33e6fd41655e372098cddc8c9480b7cd8201ca4"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2024:NUGXK6Y4YSCE545ILLRI6N2ZGR","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Scaling FP8 training to trillion-token LLMs","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Brian Chmiel, Daniel Soudry, Maxim Fishman, Ron Banner","submitted_at":"2024-09-19T07:15:58Z","abstract_excerpt":"We train, for the first time, large language models using FP8 precision on datasets up to 2 trillion tokens -- a 20-fold increase over previous limits. Through these extended training runs, we uncover critical instabilities in FP8 training that were not observable in earlier works with shorter durations. We trace these instabilities to outlier amplification by the SwiGLU activation function. Interestingly, we show, both analytically and empirically, that this amplification happens only over prolonged training periods, and link it to a SwiGLU weight alignment process. To address this newly iden"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2409.12517","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/2409.12517/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":"2026-07-05T10:11:24Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"+tc0bhpvbAF60jj+53bE5nXhSp4PSsbERufgOZ9woegmsYUapkEzjp4Hs7tqL5yTkea8kAHLXI+FGf0u3pU5DQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-03T10:58:17.207521Z"},"content_sha256":"bc9979cab3a28138a18672acc95590f8fa2ba4ae43fd2408e8cfd24bdc128ed4","schema_version":"1.0","event_id":"sha256:bc9979cab3a28138a18672acc95590f8fa2ba4ae43fd2408e8cfd24bdc128ed4"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/NUGXK6Y4YSCE545ILLRI6N2ZGR/bundle.json","state_url":"https://pith.science/pith/NUGXK6Y4YSCE545ILLRI6N2ZGR/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/NUGXK6Y4YSCE545ILLRI6N2ZGR/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":"2026-08-03T10:58:17Z","links":{"resolver":"https://pith.science/pith/NUGXK6Y4YSCE545ILLRI6N2ZGR","bundle":"https://pith.science/pith/NUGXK6Y4YSCE545ILLRI6N2ZGR/bundle.json","state":"https://pith.science/pith/NUGXK6Y4YSCE545ILLRI6N2ZGR/state.json","well_known_bundle":"https://pith.science/.well-known/pith/NUGXK6Y4YSCE545ILLRI6N2ZGR/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2024:NUGXK6Y4YSCE545ILLRI6N2ZGR","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":"9d934df82c6b51ac1b5c36f22290ff649864a57d70fa5ced999bb575b0be72aa","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2024-09-19T07:15:58Z","title_canon_sha256":"9a1a7ec3f7c6beb04465ea9b3a27caf24513566228e1ed75f298132dd59f7611"},"schema_version":"1.0","source":{"id":"2409.12517","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2409.12517","created_at":"2026-07-05T10:11:24Z"},{"alias_kind":"arxiv_version","alias_value":"2409.12517v2","created_at":"2026-07-05T10:11:24Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2409.12517","created_at":"2026-07-05T10:11:24Z"},{"alias_kind":"pith_short_12","alias_value":"NUGXK6Y4YSCE","created_at":"2026-07-05T10:11:24Z"},{"alias_kind":"pith_short_16","alias_value":"NUGXK6Y4YSCE545I","created_at":"2026-07-05T10:11:24Z"},{"alias_kind":"pith_short_8","alias_value":"NUGXK6Y4","created_at":"2026-07-05T10:11:24Z"}],"graph_snapshots":[{"event_id":"sha256:bc9979cab3a28138a18672acc95590f8fa2ba4ae43fd2408e8cfd24bdc128ed4","target":"graph","created_at":"2026-07-05T10:11:24Z","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/2409.12517/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"We train, for the first time, large language models using FP8 precision on datasets up to 2 trillion tokens -- a 20-fold increase over previous limits. Through these extended training runs, we uncover critical instabilities in FP8 training that were not observable in earlier works with shorter durations. We trace these instabilities to outlier amplification by the SwiGLU activation function. Interestingly, we show, both analytically and empirically, that this amplification happens only over prolonged training periods, and link it to a SwiGLU weight alignment process. To address this newly iden","authors_text":"Brian Chmiel, Daniel Soudry, Maxim Fishman, Ron Banner","cross_cats":["cs.AI"],"headline":"","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2024-09-19T07:15:58Z","title":"Scaling FP8 training to trillion-token LLMs"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2409.12517","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:ecd2aad0897666b20d01a97ed33e6fd41655e372098cddc8c9480b7cd8201ca4","target":"record","created_at":"2026-07-05T10:11:24Z","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":"9d934df82c6b51ac1b5c36f22290ff649864a57d70fa5ced999bb575b0be72aa","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2024-09-19T07:15:58Z","title_canon_sha256":"9a1a7ec3f7c6beb04465ea9b3a27caf24513566228e1ed75f298132dd59f7611"},"schema_version":"1.0","source":{"id":"2409.12517","kind":"arxiv","version":2}},"canonical_sha256":"6d0d757b1cc4844ef3a85ae28f37593443d6dfa2b6d9217075eb06fc2e7b7c31","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"6d0d757b1cc4844ef3a85ae28f37593443d6dfa2b6d9217075eb06fc2e7b7c31","first_computed_at":"2026-07-05T10:11:24.426732Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T10:11:24.426732Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"jrbWFqPR0aOEZc/wF1H/uG4hFNJxb1ia4+IptCU4jT5TTqkidd04qZKAxpE0SfmazvzEByF+1cc+uWRdIPx6Ag==","signature_status":"signed_v1","signed_at":"2026-07-05T10:11:24.427284Z","signed_message":"canonical_sha256_bytes"},"source_id":"2409.12517","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:ecd2aad0897666b20d01a97ed33e6fd41655e372098cddc8c9480b7cd8201ca4","sha256:bc9979cab3a28138a18672acc95590f8fa2ba4ae43fd2408e8cfd24bdc128ed4"],"state_sha256":"822a60cd0d3d9c33c1189ad6be3b5966d28a1ad39a9672f88abf459b3686a169"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Kwa8nPi2uc8NooGrhzFfnlX6uoAVkTw1itGcyuE6pesOwMlKRtGMPOqr58ufK9aLP12+p03kWKGeNJdw+DrDDw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-03T10:58:17.211390Z","bundle_sha256":"baf244ccaa9d5cfe4b8ae4a508a3aed1824a6baa70642e7d9945b03e84dc2ad1"}}