{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:AKUS4BPFZWPJMH65K4QOUCZKLE","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"4c09c0ad4d68a089a581f23dc3b02438ac5f61780a09847b115bc6d5e4a30ca7","cross_cats_sorted":["cs.AI","cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-04-20T11:47:11Z","title_canon_sha256":"27145865f7d3550908989b73d58c82ada3452e28a4997568cb9fe1166133379c"},"schema_version":"1.0","source":{"id":"2604.18128","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2604.18128","created_at":"2026-05-26T01:03:30Z"},{"alias_kind":"arxiv_version","alias_value":"2604.18128v2","created_at":"2026-05-26T01:03:30Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.18128","created_at":"2026-05-26T01:03:30Z"},{"alias_kind":"pith_short_12","alias_value":"AKUS4BPFZWPJ","created_at":"2026-05-26T01:03:30Z"},{"alias_kind":"pith_short_16","alias_value":"AKUS4BPFZWPJMH65","created_at":"2026-05-26T01:03:30Z"},{"alias_kind":"pith_short_8","alias_value":"AKUS4BPF","created_at":"2026-05-26T01:03:30Z"}],"graph_snapshots":[{"event_id":"sha256:3b0b13261c476e7e0cf64ef8f0a28d43d192e43558168e1e4963ada98a77c0e8","target":"graph","created_at":"2026-05-26T01:03:30Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","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."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","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."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","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."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","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."}],"snapshot_sha256":"58ab85e1b0d181e4b5c349833488fac87f801887f8ab34e4a25c0b609b88b776"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-20T04:24:53.568733Z","status":"completed","version":"1.0.0"}],"endpoint":"/pith/2604.18128/integrity.json","findings":[],"snapshot_sha256":"60d0cd2a3d90c58f67dbcb97c7815575e54dfaddd1b0ca01c2056042a8f477dd","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"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 ","authors_text":"Ziyang Liu","cross_cats":["cs.AI","cs.LG"],"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.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-04-20T11:47:11Z","title":"Depth Registers Unlock W4A4 on SwiGLU: A Reader/Generator Decomposition"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2604.18128","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-10T04:57:30.913578Z","id":"dc7bc49f-6ff8-4702-9fde-af6c2a5b0061","model_set":{"reader":"grok-4.3"},"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","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.","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.","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."}},"verdict_id":"dc7bc49f-6ff8-4702-9fde-af6c2a5b0061"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:8c24a0d6bc943e9ffa27fa81ea3e8df04a5137e45fa87764cf38d2cfde655d38","target":"record","created_at":"2026-05-26T01:03:30Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"4c09c0ad4d68a089a581f23dc3b02438ac5f61780a09847b115bc6d5e4a30ca7","cross_cats_sorted":["cs.AI","cs.LG"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-04-20T11:47:11Z","title_canon_sha256":"27145865f7d3550908989b73d58c82ada3452e28a4997568cb9fe1166133379c"},"schema_version":"1.0","source":{"id":"2604.18128","kind":"arxiv","version":2}},"canonical_sha256":"02a92e05e5cd9e961fdd5720ea0b2a59282cccade52d3a56e53a9e085b6c02ba","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"02a92e05e5cd9e961fdd5720ea0b2a59282cccade52d3a56e53a9e085b6c02ba","first_computed_at":"2026-05-26T01:03:30.577809Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-26T01:03:30.577809Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"7gU2m7oiSO42NTlW5v7zZPQJyHvbawNuTCI8XOigBTADV2V7B0eXFMIQv4ummygyAUrx4mXYh0TJ+0FYlFWVBg==","signature_status":"signed_v1","signed_at":"2026-05-26T01:03:30.578758Z","signed_message":"canonical_sha256_bytes"},"source_id":"2604.18128","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:8c24a0d6bc943e9ffa27fa81ea3e8df04a5137e45fa87764cf38d2cfde655d38","sha256:3b0b13261c476e7e0cf64ef8f0a28d43d192e43558168e1e4963ada98a77c0e8"],"state_sha256":"3f14c710ad78477424fb750eaf0589eaace93e0a36d3750dfcc333ec11e1fbe3"}