{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:7FZ45ZX2HLCZIBI4A3BILQ4WVP","short_pith_number":"pith:7FZ45ZX2","schema_version":"1.0","canonical_sha256":"f973cee6fa3ac594051c06c285c396abff8de4759da75db887222b9f2cfbb882","source":{"kind":"arxiv","id":"2607.02893","version":1},"attestation_state":"computed","paper":{"title":"Variable Bit-width Quantization: Learning Per-Group Precision for \"Bigger-but-Smaller\" Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Hamish Ogilvy","submitted_at":"2026-07-03T02:42:04Z","abstract_excerpt":"Low-bit quantization shrinks language models but treats precision as a single global hyper-parameter: every weight uses the same bit-width. We introduce Variable Bit-width Quantization (VBQ), a training-time method in which each contiguous group of 64 weights learns its own resolution from {1,2,4,8} bits via a Gumbel-Softmax relaxation, trained jointly by an alternating optimization that gives the precision logits a clean, task-aligned signal. VBQ discovers a consistent, strongly heterogeneous allocation within individual projection types, not merely across layers, impossible to express with p"},"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":"2607.02893","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-07-03T02:42:04Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"e16d659c78110b7d550a117c4d6fc929375e72974aa36d99d7529f53e80a63fd","abstract_canon_sha256":"a78d747900f17849e22a0457b0f8c73ffcb4e55327024ccd61ae463b61e30e8e"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-07T01:16:35.662491Z","signature_b64":"gVjJrrN2qxcpzkyl5L4kalHD23XDbOCXs6t4g9rUu7Eu/fwWmeHmhpV7lrAfUxSKaLKZqwLFACPjAYCn99QtAQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f973cee6fa3ac594051c06c285c396abff8de4759da75db887222b9f2cfbb882","last_reissued_at":"2026-07-07T01:16:35.661974Z","signature_status":"signed_v1","first_computed_at":"2026-07-07T01:16:35.661974Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Variable Bit-width Quantization: Learning Per-Group Precision for \"Bigger-but-Smaller\" Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Hamish Ogilvy","submitted_at":"2026-07-03T02:42:04Z","abstract_excerpt":"Low-bit quantization shrinks language models but treats precision as a single global hyper-parameter: every weight uses the same bit-width. We introduce Variable Bit-width Quantization (VBQ), a training-time method in which each contiguous group of 64 weights learns its own resolution from {1,2,4,8} bits via a Gumbel-Softmax relaxation, trained jointly by an alternating optimization that gives the precision logits a clean, task-aligned signal. VBQ discovers a consistent, strongly heterogeneous allocation within individual projection types, not merely across layers, impossible to express with p"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2607.02893","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/2607.02893/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":"2607.02893","created_at":"2026-07-07T01:16:35.662036+00:00"},{"alias_kind":"arxiv_version","alias_value":"2607.02893v1","created_at":"2026-07-07T01:16:35.662036+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2607.02893","created_at":"2026-07-07T01:16:35.662036+00:00"},{"alias_kind":"pith_short_12","alias_value":"7FZ45ZX2HLCZ","created_at":"2026-07-07T01:16:35.662036+00:00"},{"alias_kind":"pith_short_16","alias_value":"7FZ45ZX2HLCZIBI4","created_at":"2026-07-07T01:16:35.662036+00:00"},{"alias_kind":"pith_short_8","alias_value":"7FZ45ZX2","created_at":"2026-07-07T01:16:35.662036+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/7FZ45ZX2HLCZIBI4A3BILQ4WVP","json":"https://pith.science/pith/7FZ45ZX2HLCZIBI4A3BILQ4WVP.json","graph_json":"https://pith.science/api/pith-number/7FZ45ZX2HLCZIBI4A3BILQ4WVP/graph.json","events_json":"https://pith.science/api/pith-number/7FZ45ZX2HLCZIBI4A3BILQ4WVP/events.json","paper":"https://pith.science/paper/7FZ45ZX2"},"agent_actions":{"view_html":"https://pith.science/pith/7FZ45ZX2HLCZIBI4A3BILQ4WVP","download_json":"https://pith.science/pith/7FZ45ZX2HLCZIBI4A3BILQ4WVP.json","view_paper":"https://pith.science/paper/7FZ45ZX2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2607.02893&json=true","fetch_graph":"https://pith.science/api/pith-number/7FZ45ZX2HLCZIBI4A3BILQ4WVP/graph.json","fetch_events":"https://pith.science/api/pith-number/7FZ45ZX2HLCZIBI4A3BILQ4WVP/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7FZ45ZX2HLCZIBI4A3BILQ4WVP/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7FZ45ZX2HLCZIBI4A3BILQ4WVP/action/storage_attestation","attest_author":"https://pith.science/pith/7FZ45ZX2HLCZIBI4A3BILQ4WVP/action/author_attestation","sign_citation":"https://pith.science/pith/7FZ45ZX2HLCZIBI4A3BILQ4WVP/action/citation_signature","submit_replication":"https://pith.science/pith/7FZ45ZX2HLCZIBI4A3BILQ4WVP/action/replication_record"}},"created_at":"2026-07-07T01:16:35.662036+00:00","updated_at":"2026-07-07T01:16:35.662036+00:00"}