{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:KLPCBTFKNU4DVHBBGLFLXJ75NV","short_pith_number":"pith:KLPCBTFK","schema_version":"1.0","canonical_sha256":"52de20ccaa6d383a9c2132cabba7fd6d6d01c1ee58443c6a484d82b2a1dac359","source":{"kind":"arxiv","id":"2607.07494","version":1},"attestation_state":"computed","paper":{"title":"GIFT: Geometry-Informed Low-precision Gradient Communication for LLM Pretraining","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.DC","authors_text":"Jieying Wang, Mingkai Zheng, Shuyuan Fan, Zhao Zhang","submitted_at":"2026-07-08T14:55:51Z","abstract_excerpt":"Gradient communication is a primary scaling bottleneck in large language model (LLM) pretraining. Communicating gradients in low-precision formats, such as FP8 and NVFP4, can significantly reduce the communication volume. Existing methods quantize gradients via linear or nonlinear mappings in Euclidean space, often degrading model performance because highly anisotropic gradients incur direction-dependent distortion. We present GIFT, a geometry-informed gradient scaling method that performs low-precision communication in geometry-aware coordinates. By transforming gradients into a near-isotropi"},"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":"2607.07494","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.DC","submitted_at":"2026-07-08T14:55:51Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"c0eea2155569f925a66b4632b0f920626c61df2ce146175d06ab8bc2efbc523c","abstract_canon_sha256":"703ed529f970c438d3a2994ed275d107972a31f0c7f6ef6517956f8d77390512"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-09T01:20:27.153167Z","signature_b64":"9I9/dumnxH08gc3zuv2eYZbj/fqoyULvS0xQaygY5vYHDnP+XR87fNj/1QCFaqUM30EqR8JvVpLJOo1SPp2rAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"52de20ccaa6d383a9c2132cabba7fd6d6d01c1ee58443c6a484d82b2a1dac359","last_reissued_at":"2026-07-09T01:20:27.152734Z","signature_status":"signed_v1","first_computed_at":"2026-07-09T01:20:27.152734Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"GIFT: Geometry-Informed Low-precision Gradient Communication for LLM Pretraining","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.DC","authors_text":"Jieying Wang, Mingkai Zheng, Shuyuan Fan, Zhao Zhang","submitted_at":"2026-07-08T14:55:51Z","abstract_excerpt":"Gradient communication is a primary scaling bottleneck in large language model (LLM) pretraining. Communicating gradients in low-precision formats, such as FP8 and NVFP4, can significantly reduce the communication volume. Existing methods quantize gradients via linear or nonlinear mappings in Euclidean space, often degrading model performance because highly anisotropic gradients incur direction-dependent distortion. We present GIFT, a geometry-informed gradient scaling method that performs low-precision communication in geometry-aware coordinates. By transforming gradients into a near-isotropi"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.07494","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/2607.07494/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":"2607.07494","created_at":"2026-07-09T01:20:27.152794+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.07494v1","created_at":"2026-07-09T01:20:27.152794+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.07494","created_at":"2026-07-09T01:20:27.152794+00:00"},{"alias_kind":"pith_short_12","alias_value":"KLPCBTFKNU4D","created_at":"2026-07-09T01:20:27.152794+00:00"},{"alias_kind":"pith_short_16","alias_value":"KLPCBTFKNU4DVHBB","created_at":"2026-07-09T01:20:27.152794+00:00"},{"alias_kind":"pith_short_8","alias_value":"KLPCBTFK","created_at":"2026-07-09T01:20:27.152794+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/KLPCBTFKNU4DVHBBGLFLXJ75NV","json":"https://pith.science/pith/KLPCBTFKNU4DVHBBGLFLXJ75NV.json","graph_json":"https://pith.science/api/pith-number/KLPCBTFKNU4DVHBBGLFLXJ75NV/graph.json","events_json":"https://pith.science/api/pith-number/KLPCBTFKNU4DVHBBGLFLXJ75NV/events.json","paper":"https://pith.science/paper/KLPCBTFK"},"agent_actions":{"view_html":"https://pith.science/pith/KLPCBTFKNU4DVHBBGLFLXJ75NV","download_json":"https://pith.science/pith/KLPCBTFKNU4DVHBBGLFLXJ75NV.json","view_paper":"https://pith.science/paper/KLPCBTFK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.07494&json=true","fetch_graph":"https://pith.science/api/pith-number/KLPCBTFKNU4DVHBBGLFLXJ75NV/graph.json","fetch_events":"https://pith.science/api/pith-number/KLPCBTFKNU4DVHBBGLFLXJ75NV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/KLPCBTFKNU4DVHBBGLFLXJ75NV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/KLPCBTFKNU4DVHBBGLFLXJ75NV/action/storage_attestation","attest_author":"https://pith.science/pith/KLPCBTFKNU4DVHBBGLFLXJ75NV/action/author_attestation","sign_citation":"https://pith.science/pith/KLPCBTFKNU4DVHBBGLFLXJ75NV/action/citation_signature","submit_replication":"https://pith.science/pith/KLPCBTFKNU4DVHBBGLFLXJ75NV/action/replication_record"}},"created_at":"2026-07-09T01:20:27.152794+00:00","updated_at":"2026-07-09T01:20:27.152794+00:00"}