{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:KNP64ZBZ6XL4CYNC4HGUBXMNWZ","short_pith_number":"pith:KNP64ZBZ","canonical_record":{"source":{"id":"2605.10886","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-11T17:32:29Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"4751061f62e705cdf6a3b9d3f4a74c5f44efd30e84183a009b3d30bc05e82554","abstract_canon_sha256":"5be490cd6aca8747f8d9229bc46a30a38d0f817cd17a79206fac2e03d256972a"},"schema_version":"1.0"},"canonical_sha256":"535fee6439f5d7c161a2e1cd40dd8db6540e05c660eff1da1c1daa67aa52e3ae","source":{"kind":"arxiv","id":"2605.10886","version":3},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.10886","created_at":"2026-07-10T00:18:45Z"},{"alias_kind":"arxiv_version","alias_value":"2605.10886v3","created_at":"2026-07-10T00:18:45Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.10886","created_at":"2026-07-10T00:18:45Z"},{"alias_kind":"pith_short_12","alias_value":"KNP64ZBZ6XL4","created_at":"2026-07-10T00:18:45Z"},{"alias_kind":"pith_short_16","alias_value":"KNP64ZBZ6XL4CYNC","created_at":"2026-07-10T00:18:45Z"},{"alias_kind":"pith_short_8","alias_value":"KNP64ZBZ","created_at":"2026-07-10T00:18:45Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:KNP64ZBZ6XL4CYNC4HGUBXMNWZ","target":"record","payload":{"canonical_record":{"source":{"id":"2605.10886","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-11T17:32:29Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"4751061f62e705cdf6a3b9d3f4a74c5f44efd30e84183a009b3d30bc05e82554","abstract_canon_sha256":"5be490cd6aca8747f8d9229bc46a30a38d0f817cd17a79206fac2e03d256972a"},"schema_version":"1.0"},"canonical_sha256":"535fee6439f5d7c161a2e1cd40dd8db6540e05c660eff1da1c1daa67aa52e3ae","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-10T00:18:45.426466Z","signature_b64":"GKOedE6UyLg/HBOX0YdNOmEhQb24dIMgRSGO5txx9rLwTa6KuO1HUWBD7oeJ8Wt4Bos4ULhvO842vN5VCEsSBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"535fee6439f5d7c161a2e1cd40dd8db6540e05c660eff1da1c1daa67aa52e3ae","last_reissued_at":"2026-07-10T00:18:45.425954Z","signature_status":"signed_v1","first_computed_at":"2026-07-10T00:18:45.425954Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.10886","source_version":3,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-10T00:18:45Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"gGI9hdezEga0nCXiJCGzrUfpQtfiCRVPBvPMKTTFffXgI7MmmUNzXwZMvfNcdUTNFzeAWnF3FYZfwaUbqRwJCg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-03T18:11:40.007546Z"},"content_sha256":"1942baa26e27ff09bd2500ba6dab0f091e89b062673f507fbe2531047bbc7611","schema_version":"1.0","event_id":"sha256:1942baa26e27ff09bd2500ba6dab0f091e89b062673f507fbe2531047bbc7611"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:KNP64ZBZ6XL4CYNC4HGUBXMNWZ","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"LoKA: Low-precision Kernel Applications for Recommendation Models At Scale","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"LoKA framework makes FP8 practical for large recommendation models by profiling safe sites, adapting components, and dispatching kernels.","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Buyun Zhang, Chunqiang Tang, Chunzhi Yang, Ellie Wen, Jian Jiao, Jiecao Yu, Liang Luo, Maxim Naumov, Neng Shi, Quanyu Zhu, Sandeep Parab, Santanu Kolay, Shen Li, Tongyi Tang, Vasiliy Kuznetsov, Venkatesh Ranganathan, Wenlin Chen, Xiaohan Wei, Yanli Zhao, Yantao Yao, Yinbin Ma, Yuchen Hao, Yuxin Chen, Zeliang Chen","submitted_at":"2026-05-11T17:32:29Z","abstract_excerpt":"Recent GPU generations deliver significantly higher FLOPs using lower-precision arithmetic, such as FP8. While successfully applied to large language models (LLMs), its adoption in large recommendation models (LRMs) has been limited. This is because LRMs are numerically sensitive, dominated by small matrix multiplications (GEMMs) followed by normalization, and trained in communication-intensive environments. Applying FP8 directly to LRMs often degrades model quality and prolongs training time. These challenges are inherent to LRM workloads and cannot be resolved merely by introducing better FP"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"LoKA makes FP8 practical for LRMs through three principles: profile under realistic distributions to know where low precision is safe, co-design model components with hardware to expand where it is safe, and orchestrate across kernel libraries to maximize the gains.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the statistical profiling from LoKA Probe accurately identifies all safe FP8 sites without missing interactions or distribution shifts that would degrade overall model quality during full training.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"LoKA enables practical FP8 use in numerically sensitive large recommendation models via online profiling of activations, reusable model modifications for stability, and dynamic kernel dispatching.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"LoKA framework makes FP8 practical for large recommendation models by profiling safe sites, adapting components, and dispatching kernels.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"42d31b4456c6ba02cf3fc4299034c597bd51126b3af6ef731e25f7771647a076"},"source":{"id":"2605.10886","kind":"arxiv","version":3},"verdict":{"id":"59b068c0-129a-4974-b40f-51495c6cbe0f","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T04:52:02.893368Z","strongest_claim":"LoKA makes FP8 practical for LRMs through three principles: profile under realistic distributions to know where low precision is safe, co-design model components with hardware to expand where it is safe, and orchestrate across kernel libraries to maximize the gains.","one_line_summary":"LoKA enables practical FP8 use in numerically sensitive large recommendation models via online profiling of activations, reusable model modifications for stability, and dynamic kernel dispatching.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the statistical profiling from LoKA Probe accurately identifies all safe FP8 sites without missing interactions or distribution shifts that would degrade overall model quality during full training.","pith_extraction_headline":"LoKA framework makes FP8 practical for large recommendation models by profiling safe sites, adapting components, and dispatching kernels."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.10886/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-20T05:02:01.011936Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T14:33:50.963952Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T10:31:17.368796Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T08:55:47.428013Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"2bf14d82592325f1b93075337894b955edf264ddb55b61482326e7218292b4ea"},"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":"59b068c0-129a-4974-b40f-51495c6cbe0f"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-10T00:18:45Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"EFxjUIIbrueNPbA1FKpAT65PTbRd/oJSKU++eI5TiNiY+c6sUqRFpQa8mrZ18Dai/fcgseCm2CRLv2hJTATkBw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-03T18:11:40.008132Z"},"content_sha256":"b0fca2afc20d7c0600dae59755e943b21b5e5fa912f47022b16ec131c38d284c","schema_version":"1.0","event_id":"sha256:b0fca2afc20d7c0600dae59755e943b21b5e5fa912f47022b16ec131c38d284c"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/KNP64ZBZ6XL4CYNC4HGUBXMNWZ/bundle.json","state_url":"https://pith.science/pith/KNP64ZBZ6XL4CYNC4HGUBXMNWZ/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/KNP64ZBZ6XL4CYNC4HGUBXMNWZ/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":"2026-08-03T18:11:40Z","links":{"resolver":"https://pith.science/pith/KNP64ZBZ6XL4CYNC4HGUBXMNWZ","bundle":"https://pith.science/pith/KNP64ZBZ6XL4CYNC4HGUBXMNWZ/bundle.json","state":"https://pith.science/pith/KNP64ZBZ6XL4CYNC4HGUBXMNWZ/state.json","well_known_bundle":"https://pith.science/.well-known/pith/KNP64ZBZ6XL4CYNC4HGUBXMNWZ/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:KNP64ZBZ6XL4CYNC4HGUBXMNWZ","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":"5be490cd6aca8747f8d9229bc46a30a38d0f817cd17a79206fac2e03d256972a","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-11T17:32:29Z","title_canon_sha256":"4751061f62e705cdf6a3b9d3f4a74c5f44efd30e84183a009b3d30bc05e82554"},"schema_version":"1.0","source":{"id":"2605.10886","kind":"arxiv","version":3}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.10886","created_at":"2026-07-10T00:18:45Z"},{"alias_kind":"arxiv_version","alias_value":"2605.10886v3","created_at":"2026-07-10T00:18:45Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.10886","created_at":"2026-07-10T00:18:45Z"},{"alias_kind":"pith_short_12","alias_value":"KNP64ZBZ6XL4","created_at":"2026-07-10T00:18:45Z"},{"alias_kind":"pith_short_16","alias_value":"KNP64ZBZ6XL4CYNC","created_at":"2026-07-10T00:18:45Z"},{"alias_kind":"pith_short_8","alias_value":"KNP64ZBZ","created_at":"2026-07-10T00:18:45Z"}],"graph_snapshots":[{"event_id":"sha256:b0fca2afc20d7c0600dae59755e943b21b5e5fa912f47022b16ec131c38d284c","target":"graph","created_at":"2026-07-10T00:18:45Z","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":"LoKA makes FP8 practical for LRMs through three principles: profile under realistic distributions to know where low precision is safe, co-design model components with hardware to expand where it is safe, and orchestrate across kernel libraries to maximize the gains."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the statistical profiling from LoKA Probe accurately identifies all safe FP8 sites without missing interactions or distribution shifts that would degrade overall model quality during full training."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"LoKA enables practical FP8 use in numerically sensitive large recommendation models via online profiling of activations, reusable model modifications for stability, and dynamic kernel dispatching."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"LoKA framework makes FP8 practical for large recommendation models by profiling safe sites, adapting components, and dispatching kernels."}],"snapshot_sha256":"42d31b4456c6ba02cf3fc4299034c597bd51126b3af6ef731e25f7771647a076"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"claim_evidence","ran_at":"2026-05-20T05:02:01.011936Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-19T14:33:50.963952Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_title_agreement","ran_at":"2026-05-19T10:31:17.368796Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-19T08:55:47.428013Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2605.10886/integrity.json","findings":[],"snapshot_sha256":"2bf14d82592325f1b93075337894b955edf264ddb55b61482326e7218292b4ea","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Recent GPU generations deliver significantly higher FLOPs using lower-precision arithmetic, such as FP8. While successfully applied to large language models (LLMs), its adoption in large recommendation models (LRMs) has been limited. This is because LRMs are numerically sensitive, dominated by small matrix multiplications (GEMMs) followed by normalization, and trained in communication-intensive environments. Applying FP8 directly to LRMs often degrades model quality and prolongs training time. These challenges are inherent to LRM workloads and cannot be resolved merely by introducing better FP","authors_text":"Buyun Zhang, Chunqiang Tang, Chunzhi Yang, Ellie Wen, Jian Jiao, Jiecao Yu, Liang Luo, Maxim Naumov, Neng Shi, Quanyu Zhu, Sandeep Parab, Santanu Kolay, Shen Li, Tongyi Tang, Vasiliy Kuznetsov, Venkatesh Ranganathan, Wenlin Chen, Xiaohan Wei, Yanli Zhao, Yantao Yao, Yinbin Ma, Yuchen Hao, Yuxin Chen, Zeliang Chen","cross_cats":["cs.AI"],"headline":"LoKA framework makes FP8 practical for large recommendation models by profiling safe sites, adapting components, and dispatching kernels.","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-11T17:32:29Z","title":"LoKA: Low-precision Kernel Applications for Recommendation Models At Scale"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2605.10886","kind":"arxiv","version":3},"verdict":{"created_at":"2026-05-15T04:52:02.893368Z","id":"59b068c0-129a-4974-b40f-51495c6cbe0f","model_set":{"reader":"grok-4.3"},"one_line_summary":"LoKA enables practical FP8 use in numerically sensitive large recommendation models via online profiling of activations, reusable model modifications for stability, and dynamic kernel dispatching.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"LoKA framework makes FP8 practical for large recommendation models by profiling safe sites, adapting components, and dispatching kernels.","strongest_claim":"LoKA makes FP8 practical for LRMs through three principles: profile under realistic distributions to know where low precision is safe, co-design model components with hardware to expand where it is safe, and orchestrate across kernel libraries to maximize the gains.","weakest_assumption":"That the statistical profiling from LoKA Probe accurately identifies all safe FP8 sites without missing interactions or distribution shifts that would degrade overall model quality during full training."}},"verdict_id":"59b068c0-129a-4974-b40f-51495c6cbe0f"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:1942baa26e27ff09bd2500ba6dab0f091e89b062673f507fbe2531047bbc7611","target":"record","created_at":"2026-07-10T00:18:45Z","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":"5be490cd6aca8747f8d9229bc46a30a38d0f817cd17a79206fac2e03d256972a","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-11T17:32:29Z","title_canon_sha256":"4751061f62e705cdf6a3b9d3f4a74c5f44efd30e84183a009b3d30bc05e82554"},"schema_version":"1.0","source":{"id":"2605.10886","kind":"arxiv","version":3}},"canonical_sha256":"535fee6439f5d7c161a2e1cd40dd8db6540e05c660eff1da1c1daa67aa52e3ae","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"535fee6439f5d7c161a2e1cd40dd8db6540e05c660eff1da1c1daa67aa52e3ae","first_computed_at":"2026-07-10T00:18:45.425954Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-10T00:18:45.425954Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"GKOedE6UyLg/HBOX0YdNOmEhQb24dIMgRSGO5txx9rLwTa6KuO1HUWBD7oeJ8Wt4Bos4ULhvO842vN5VCEsSBw==","signature_status":"signed_v1","signed_at":"2026-07-10T00:18:45.426466Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.10886","source_kind":"arxiv","source_version":3}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:1942baa26e27ff09bd2500ba6dab0f091e89b062673f507fbe2531047bbc7611","sha256:b0fca2afc20d7c0600dae59755e943b21b5e5fa912f47022b16ec131c38d284c"],"state_sha256":"00d90ee15dfe466bfd9769a4a02731a3c721201d5b08dfd6c3fd2ed942ccccf1"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"90bjy2zHFOcROA6h/HXM+wTCeBEEgFuHyWFqxFYYXccw77bhdfNQpAvJwRDzogMWilMZsCXCr8xbkGMN1rNaCA==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-03T18:11:40.014023Z","bundle_sha256":"a713ccf1fa6d206d3291661679376ddabaab7b56d1dc4ea9d31d11a8429a696c"}}