{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:AO4RZQ252RSMMEBLJTSU4MN3JI","short_pith_number":"pith:AO4RZQ25","schema_version":"1.0","canonical_sha256":"03b91cc35dd464c6102b4ce54e31bb4a0c6da75a56dd92efa7d70e24476f1870","source":{"kind":"arxiv","id":"2608.12026","version":1},"attestation_state":"computed","paper":{"title":"SoftWater: Class-Aware Rate Allocation for Softmax Quantization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Ashia C. Wilson, Joao V. Cavalcanti","submitted_at":"2026-08-12T13:06:50Z","abstract_excerpt":"Post-training quantization pipelines routinely leave the softmax output layer in high precision. Yet in small LLMs with modern vocabularies, the head holds 15--30\\% of all parameters, so a nominal ``2-bit'' model with an fp16 head can store several times as many bits per weight. We pose softmax-layer quantization as a rate-distortion problem under the KL divergence between the original and quantized output distributions. A second-order analysis reveals a class-aware geometry: quantization error is weighted jointly by feature covariance and class-specific softmax curvature. A separability appro"},"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":"2608.12026","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-08-12T13:06:50Z","cross_cats_sorted":[],"title_canon_sha256":"db80b3ca354e209c39cbaaec1af9192679f49712c42b375326e3e9a579fc6d8d","abstract_canon_sha256":"0b60cae7919e8ae0ec65c4d326c57c1722426f71d5655b26cdda9c99395d2250"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-13T01:29:17.103312Z","signature_b64":"rJapqW+RypwHzEwqLLQz98jU2U3zgVsjkZ6Ou165ZGeb1JyF1BwzDDQEdX44RZ0WBC/Sv1sBrr3s1sfUiLD9Aw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"03b91cc35dd464c6102b4ce54e31bb4a0c6da75a56dd92efa7d70e24476f1870","last_reissued_at":"2026-08-13T01:29:17.101341Z","signature_status":"signed_v1","first_computed_at":"2026-08-13T01:29:17.101341Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"SoftWater: Class-Aware Rate Allocation for Softmax Quantization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Ashia C. Wilson, Joao V. Cavalcanti","submitted_at":"2026-08-12T13:06:50Z","abstract_excerpt":"Post-training quantization pipelines routinely leave the softmax output layer in high precision. Yet in small LLMs with modern vocabularies, the head holds 15--30\\% of all parameters, so a nominal ``2-bit'' model with an fp16 head can store several times as many bits per weight. We pose softmax-layer quantization as a rate-distortion problem under the KL divergence between the original and quantized output distributions. A second-order analysis reveals a class-aware geometry: quantization error is weighted jointly by feature covariance and class-specific softmax curvature. A separability appro"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2608.12026","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/2608.12026/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":"2608.12026","created_at":"2026-08-13T01:29:17.102359+00:00"},{"alias_kind":"arxiv_version","alias_value":"2608.12026v1","created_at":"2026-08-13T01:29:17.102359+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2608.12026","created_at":"2026-08-13T01:29:17.102359+00:00"},{"alias_kind":"pith_short_12","alias_value":"AO4RZQ252RSM","created_at":"2026-08-13T01:29:17.102359+00:00"},{"alias_kind":"pith_short_16","alias_value":"AO4RZQ252RSMMEBL","created_at":"2026-08-13T01:29:17.102359+00:00"},{"alias_kind":"pith_short_8","alias_value":"AO4RZQ25","created_at":"2026-08-13T01:29:17.102359+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/AO4RZQ252RSMMEBLJTSU4MN3JI","json":"https://pith.science/pith/AO4RZQ252RSMMEBLJTSU4MN3JI.json","graph_json":"https://pith.science/api/pith-number/AO4RZQ252RSMMEBLJTSU4MN3JI/graph.json","events_json":"https://pith.science/api/pith-number/AO4RZQ252RSMMEBLJTSU4MN3JI/events.json","paper":"https://pith.science/paper/AO4RZQ25"},"agent_actions":{"view_html":"https://pith.science/pith/AO4RZQ252RSMMEBLJTSU4MN3JI","download_json":"https://pith.science/pith/AO4RZQ252RSMMEBLJTSU4MN3JI.json","view_paper":"https://pith.science/paper/AO4RZQ25","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2608.12026&json=true","fetch_graph":"https://pith.science/api/pith-number/AO4RZQ252RSMMEBLJTSU4MN3JI/graph.json","fetch_events":"https://pith.science/api/pith-number/AO4RZQ252RSMMEBLJTSU4MN3JI/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/AO4RZQ252RSMMEBLJTSU4MN3JI/action/timestamp_anchor","attest_storage":"https://pith.science/pith/AO4RZQ252RSMMEBLJTSU4MN3JI/action/storage_attestation","attest_author":"https://pith.science/pith/AO4RZQ252RSMMEBLJTSU4MN3JI/action/author_attestation","sign_citation":"https://pith.science/pith/AO4RZQ252RSMMEBLJTSU4MN3JI/action/citation_signature","submit_replication":"https://pith.science/pith/AO4RZQ252RSMMEBLJTSU4MN3JI/action/replication_record"}},"created_at":"2026-08-13T01:29:17.102359+00:00","updated_at":"2026-08-13T01:29:17.102359+00:00"}