{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:AKUS4BPFZWPJMH65K4QOUCZKLE","short_pith_number":"pith:AKUS4BPF","schema_version":"1.0","canonical_sha256":"02a92e05e5cd9e961fdd5720ea0b2a59282cccade52d3a56e53a9e085b6c02ba","source":{"kind":"arxiv","id":"2604.18128","version":2},"attestation_state":"computed","paper":{"title":"Depth Registers Unlock W4A4 on SwiGLU: A Reader/Generator Decomposition","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Depth registers with a hinge loss control residual reader magnitudes in SwiGLU, unlocking W4A4 quantization performance while leaving the w2 bilinear generator as the dominant error.","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Ziyang Liu","submitted_at":"2026-04-20T11:47:11Z","abstract_excerpt":"We study post-training W4A4 quantization in a controlled 300M-parameter SwiGLU decoder-only language model trained on 5B tokens of FineWeb-Edu, and ask which input-activation sites dominate the error. Naive round-to-nearest W4A4 collapses validation perplexity from FP16 23.6 to 1727. A simple residual-axis training-time intervention -- Depth Registers with a register-magnitude hinge loss (DR+sink) -- reduces this to 119 (about 14x) at matched FP16 PPL and matched zero-shot capacity, and composes with SmoothQuant to 39.9 PPL. The residual ~2 PPL gap to FP16 is the diagnostic core. We decompose "},"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":"2604.18128","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-04-20T11:47:11Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"27145865f7d3550908989b73d58c82ada3452e28a4997568cb9fe1166133379c","abstract_canon_sha256":"4c09c0ad4d68a089a581f23dc3b02438ac5f61780a09847b115bc6d5e4a30ca7"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-26T01:03:30.578758Z","signature_b64":"7gU2m7oiSO42NTlW5v7zZPQJyHvbawNuTCI8XOigBTADV2V7B0eXFMIQv4ummygyAUrx4mXYh0TJ+0FYlFWVBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"02a92e05e5cd9e961fdd5720ea0b2a59282cccade52d3a56e53a9e085b6c02ba","last_reissued_at":"2026-05-26T01:03:30.577809Z","signature_status":"signed_v1","first_computed_at":"2026-05-26T01:03:30.577809Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Depth Registers Unlock W4A4 on SwiGLU: A Reader/Generator Decomposition","license":"http://creativecommons.org/licenses/by/4.0/","headline":"Depth registers with a hinge loss control residual reader magnitudes in SwiGLU, unlocking W4A4 quantization performance while leaving the w2 bilinear generator as the dominant error.","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Ziyang Liu","submitted_at":"2026-04-20T11:47:11Z","abstract_excerpt":"We study post-training W4A4 quantization in a controlled 300M-parameter SwiGLU decoder-only language model trained on 5B tokens of FineWeb-Edu, and ask which input-activation sites dominate the error. Naive round-to-nearest W4A4 collapses validation perplexity from FP16 23.6 to 1727. A simple residual-axis training-time intervention -- Depth Registers with a register-magnitude hinge loss (DR+sink) -- reduces this to 119 (about 14x) at matched FP16 PPL and matched zero-shot capacity, and composes with SmoothQuant to 39.9 PPL. The residual ~2 PPL gap to FP16 is the diagnostic core. We decompose "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Elementary norm arguments show residual-axis magnitude control bounds readers tightly but leaves w2's bilinear input bounded only by the trivial product of factor bounds; empirically, DR+sink collapses reader kurtosis while leaving generators essentially unchanged, and the reader-rescued W4A4 residue is flat at ~0.28 nats across three matched checkpoints with Delta-remove(w2) dominating.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The partition into residual-axis readers (qkv, w1, w3) and block-internal generators (o_proj, w2) plus the claim that the hinge loss isolates the reader effect without confounding the generator behavior; the paper states its experiments do not isolate the partition from the hinge.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Depth Registers plus hinge loss cut W4A4-induced perplexity collapse from 1727 to 119 in a 300M SwiGLU model by selectively taming reader-layer activations while leaving bilinear generator tails largely untouched.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Depth registers with a hinge loss control residual reader magnitudes in SwiGLU, unlocking W4A4 quantization performance while leaving the w2 bilinear generator as the dominant error.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"58ab85e1b0d181e4b5c349833488fac87f801887f8ab34e4a25c0b609b88b776"},"source":{"id":"2604.18128","kind":"arxiv","version":2},"verdict":{"id":"dc7bc49f-6ff8-4702-9fde-af6c2a5b0061","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T04:57:30.913578Z","strongest_claim":"Elementary norm arguments show residual-axis magnitude control bounds readers tightly but leaves w2's bilinear input bounded only by the trivial product of factor bounds; empirically, DR+sink collapses reader kurtosis while leaving generators essentially unchanged, and the reader-rescued W4A4 residue is flat at ~0.28 nats across three matched checkpoints with Delta-remove(w2) dominating.","one_line_summary":"Depth Registers plus hinge loss cut W4A4-induced perplexity collapse from 1727 to 119 in a 300M SwiGLU model by selectively taming reader-layer activations while leaving bilinear generator tails largely untouched.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The partition into residual-axis readers (qkv, w1, w3) and block-internal generators (o_proj, w2) plus the claim that the hinge loss isolates the reader effect without confounding the generator behavior; the paper states its experiments do not isolate the partition from the hinge.","pith_extraction_headline":"Depth registers with a hinge loss control residual reader magnitudes in SwiGLU, unlocking W4A4 quantization performance while leaving the w2 bilinear generator as the dominant error."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.18128/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"doi_compliance","ran_at":"2026-05-20T04:24:53.568733Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"60d0cd2a3d90c58f67dbcb97c7815575e54dfaddd1b0ca01c2056042a8f477dd"},"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":"2604.18128","created_at":"2026-05-26T01:03:30.577956+00:00"},{"alias_kind":"arxiv_version","alias_value":"2604.18128v2","created_at":"2026-05-26T01:03:30.577956+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.18128","created_at":"2026-05-26T01:03:30.577956+00:00"},{"alias_kind":"pith_short_12","alias_value":"AKUS4BPFZWPJ","created_at":"2026-05-26T01:03:30.577956+00:00"},{"alias_kind":"pith_short_16","alias_value":"AKUS4BPFZWPJMH65","created_at":"2026-05-26T01:03:30.577956+00:00"},{"alias_kind":"pith_short_8","alias_value":"AKUS4BPF","created_at":"2026-05-26T01:03:30.577956+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/AKUS4BPFZWPJMH65K4QOUCZKLE","json":"https://pith.science/pith/AKUS4BPFZWPJMH65K4QOUCZKLE.json","graph_json":"https://pith.science/api/pith-number/AKUS4BPFZWPJMH65K4QOUCZKLE/graph.json","events_json":"https://pith.science/api/pith-number/AKUS4BPFZWPJMH65K4QOUCZKLE/events.json","paper":"https://pith.science/paper/AKUS4BPF"},"agent_actions":{"view_html":"https://pith.science/pith/AKUS4BPFZWPJMH65K4QOUCZKLE","download_json":"https://pith.science/pith/AKUS4BPFZWPJMH65K4QOUCZKLE.json","view_paper":"https://pith.science/paper/AKUS4BPF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2604.18128&json=true","fetch_graph":"https://pith.science/api/pith-number/AKUS4BPFZWPJMH65K4QOUCZKLE/graph.json","fetch_events":"https://pith.science/api/pith-number/AKUS4BPFZWPJMH65K4QOUCZKLE/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/AKUS4BPFZWPJMH65K4QOUCZKLE/action/timestamp_anchor","attest_storage":"https://pith.science/pith/AKUS4BPFZWPJMH65K4QOUCZKLE/action/storage_attestation","attest_author":"https://pith.science/pith/AKUS4BPFZWPJMH65K4QOUCZKLE/action/author_attestation","sign_citation":"https://pith.science/pith/AKUS4BPFZWPJMH65K4QOUCZKLE/action/citation_signature","submit_replication":"https://pith.science/pith/AKUS4BPFZWPJMH65K4QOUCZKLE/action/replication_record"}},"created_at":"2026-05-26T01:03:30.577956+00:00","updated_at":"2026-05-26T01:03:30.577956+00:00"}