{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:ET7PT6SRPZ7B5ECVE4BUZDCYZQ","short_pith_number":"pith:ET7PT6SR","schema_version":"1.0","canonical_sha256":"24fef9fa517e7e1e905527034c8c58cc38a998fbac0bae017f5157f93f3102df","source":{"kind":"arxiv","id":"2607.15105","version":1},"attestation_state":"computed","paper":{"title":"Long-Context Fine-Tuning with Limited VRAM","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Aleksandr Sazhin, Artemiy Grinenko, Frank Woernle, Vladimir Fedosov","submitted_at":"2026-07-16T15:16:06Z","abstract_excerpt":"Parameter-efficient fine-tuning reduces model and optimizer memory, but dense attention still makes long training sequences expensive. We combine Hierarchical Global Attention (HGA) with segment-wise backpropagation and tiered KV storage. Only the active segment remains differentiable in VRAM; older KV is detached into RAM or NVMe, and HGA loads a bounded set of exact historical tokens for each query block. On Qwen3-8B with 4-bit QLoRA and PG19, dense training on a 16 GB Quadro RTX 5000 fits 2,048 tokens but fails at 4,096, whereas HGA reaches 16,384 tokens with 15.28 GB peak VRAM. Under evalu"},"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":"2607.15105","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AI","submitted_at":"2026-07-16T15:16:06Z","cross_cats_sorted":[],"title_canon_sha256":"79269b1a28fc3b7fdb437492e46374bf4f68d70c396fd5da0d6d4176d02f9129","abstract_canon_sha256":"723a9ec45070f7aee636640dabca4f10a19c8f4a4ae80cbfe93cf905eea433a4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-17T01:22:09.690573Z","signature_b64":"3/+C2/2AhHze0gfc82aJ1C5u/eZ7lnbtBaXMQQ5ucRj28ABuzSf/g/irkiBTlJWVTUXIfYI5PRtDLBdvbetaAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"24fef9fa517e7e1e905527034c8c58cc38a998fbac0bae017f5157f93f3102df","last_reissued_at":"2026-07-17T01:22:09.689764Z","signature_status":"signed_v1","first_computed_at":"2026-07-17T01:22:09.689764Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Long-Context Fine-Tuning with Limited VRAM","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Aleksandr Sazhin, Artemiy Grinenko, Frank Woernle, Vladimir Fedosov","submitted_at":"2026-07-16T15:16:06Z","abstract_excerpt":"Parameter-efficient fine-tuning reduces model and optimizer memory, but dense attention still makes long training sequences expensive. We combine Hierarchical Global Attention (HGA) with segment-wise backpropagation and tiered KV storage. Only the active segment remains differentiable in VRAM; older KV is detached into RAM or NVMe, and HGA loads a bounded set of exact historical tokens for each query block. On Qwen3-8B with 4-bit QLoRA and PG19, dense training on a 16 GB Quadro RTX 5000 fits 2,048 tokens but fails at 4,096, whereas HGA reaches 16,384 tokens with 15.28 GB peak VRAM. Under evalu"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.15105","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/2607.15105/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":"2607.15105","created_at":"2026-07-17T01:22:09.690188+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.15105v1","created_at":"2026-07-17T01:22:09.690188+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.15105","created_at":"2026-07-17T01:22:09.690188+00:00"},{"alias_kind":"pith_short_12","alias_value":"ET7PT6SRPZ7B","created_at":"2026-07-17T01:22:09.690188+00:00"},{"alias_kind":"pith_short_16","alias_value":"ET7PT6SRPZ7B5ECV","created_at":"2026-07-17T01:22:09.690188+00:00"},{"alias_kind":"pith_short_8","alias_value":"ET7PT6SR","created_at":"2026-07-17T01:22:09.690188+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/ET7PT6SRPZ7B5ECVE4BUZDCYZQ","json":"https://pith.science/pith/ET7PT6SRPZ7B5ECVE4BUZDCYZQ.json","graph_json":"https://pith.science/api/pith-number/ET7PT6SRPZ7B5ECVE4BUZDCYZQ/graph.json","events_json":"https://pith.science/api/pith-number/ET7PT6SRPZ7B5ECVE4BUZDCYZQ/events.json","paper":"https://pith.science/paper/ET7PT6SR"},"agent_actions":{"view_html":"https://pith.science/pith/ET7PT6SRPZ7B5ECVE4BUZDCYZQ","download_json":"https://pith.science/pith/ET7PT6SRPZ7B5ECVE4BUZDCYZQ.json","view_paper":"https://pith.science/paper/ET7PT6SR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.15105&json=true","fetch_graph":"https://pith.science/api/pith-number/ET7PT6SRPZ7B5ECVE4BUZDCYZQ/graph.json","fetch_events":"https://pith.science/api/pith-number/ET7PT6SRPZ7B5ECVE4BUZDCYZQ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ET7PT6SRPZ7B5ECVE4BUZDCYZQ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ET7PT6SRPZ7B5ECVE4BUZDCYZQ/action/storage_attestation","attest_author":"https://pith.science/pith/ET7PT6SRPZ7B5ECVE4BUZDCYZQ/action/author_attestation","sign_citation":"https://pith.science/pith/ET7PT6SRPZ7B5ECVE4BUZDCYZQ/action/citation_signature","submit_replication":"https://pith.science/pith/ET7PT6SRPZ7B5ECVE4BUZDCYZQ/action/replication_record"}},"created_at":"2026-07-17T01:22:09.690188+00:00","updated_at":"2026-07-17T01:22:09.690188+00:00"}