{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:SZIRWIBGDCDKG6ONJHRIASHKA5","short_pith_number":"pith:SZIRWIBG","schema_version":"1.0","canonical_sha256":"96511b20261886a379cd49e28048ea074c119b7007d28b6873974abf8f2b3eb1","source":{"kind":"arxiv","id":"2608.02689","version":1},"attestation_state":"computed","paper":{"title":"Stuck on \"A\": Diagnosing and Repairing Interface Injury in Attention-to-KDA Linearization of a 0.6B Language Model","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Ronglong Bao","submitted_at":"2026-08-03T08:54:18Z","abstract_excerpt":"We convert 21 of 28 full-attention layers of Qwen3-0.6B-Base into KDA (Kimi Delta Attention) linear-attention layers on a single consumer-grade GPU budget, and ask a simple question: what exactly does the conversion break? After surgery, hidden-state alignment and end-to-end KL distillation drive the student close to its teacher in perplexity, yet multiple-choice accuracy stays near random chance (25-29% vs. the teacher's 50.6% on C-Eval). Using a four-permutation diagnostic that rotates answer options while holding content fixed, we show the model sticks to option labels (predicting \"A\" 81% o"},"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":"2608.02689","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2026-08-03T08:54:18Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"6e1054752ac3ad87e7b468065904195b8e76241b3159b60fa4599ba5402f65f4","abstract_canon_sha256":"508710153e7fbcc45a734d870fdbf5a62735ad60ae07fe5132fb0e53e86f4f7f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-08-05T00:39:25.117987Z","signature_b64":"K/sEJIy8+s6JkwRa4pzNfkg8GEpyngghrfN1zm89YwX1kSm70YrZIufL41GjUOkTDYnWxXo6b/mcJmZ/qLRfCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"96511b20261886a379cd49e28048ea074c119b7007d28b6873974abf8f2b3eb1","last_reissued_at":"2026-08-05T00:39:25.115467Z","signature_status":"signed_v1","first_computed_at":"2026-08-05T00:39:25.115467Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Stuck on \"A\": Diagnosing and Repairing Interface Injury in Attention-to-KDA Linearization of a 0.6B Language Model","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Ronglong Bao","submitted_at":"2026-08-03T08:54:18Z","abstract_excerpt":"We convert 21 of 28 full-attention layers of Qwen3-0.6B-Base into KDA (Kimi Delta Attention) linear-attention layers on a single consumer-grade GPU budget, and ask a simple question: what exactly does the conversion break? After surgery, hidden-state alignment and end-to-end KL distillation drive the student close to its teacher in perplexity, yet multiple-choice accuracy stays near random chance (25-29% vs. the teacher's 50.6% on C-Eval). Using a four-permutation diagnostic that rotates answer options while holding content fixed, we show the model sticks to option labels (predicting \"A\" 81% o"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2608.02689","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/2608.02689/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":"2608.02689","created_at":"2026-08-05T00:39:25.116328+00:00"},{"alias_kind":"arxiv_version","alias_value":"2608.02689v1","created_at":"2026-08-05T00:39:25.116328+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2608.02689","created_at":"2026-08-05T00:39:25.116328+00:00"},{"alias_kind":"pith_short_12","alias_value":"SZIRWIBGDCDK","created_at":"2026-08-05T00:39:25.116328+00:00"},{"alias_kind":"pith_short_16","alias_value":"SZIRWIBGDCDKG6ON","created_at":"2026-08-05T00:39:25.116328+00:00"},{"alias_kind":"pith_short_8","alias_value":"SZIRWIBG","created_at":"2026-08-05T00:39:25.116328+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/SZIRWIBGDCDKG6ONJHRIASHKA5","json":"https://pith.science/pith/SZIRWIBGDCDKG6ONJHRIASHKA5.json","graph_json":"https://pith.science/api/pith-number/SZIRWIBGDCDKG6ONJHRIASHKA5/graph.json","events_json":"https://pith.science/api/pith-number/SZIRWIBGDCDKG6ONJHRIASHKA5/events.json","paper":"https://pith.science/paper/SZIRWIBG"},"agent_actions":{"view_html":"https://pith.science/pith/SZIRWIBGDCDKG6ONJHRIASHKA5","download_json":"https://pith.science/pith/SZIRWIBGDCDKG6ONJHRIASHKA5.json","view_paper":"https://pith.science/paper/SZIRWIBG","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2608.02689&json=true","fetch_graph":"https://pith.science/api/pith-number/SZIRWIBGDCDKG6ONJHRIASHKA5/graph.json","fetch_events":"https://pith.science/api/pith-number/SZIRWIBGDCDKG6ONJHRIASHKA5/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SZIRWIBGDCDKG6ONJHRIASHKA5/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SZIRWIBGDCDKG6ONJHRIASHKA5/action/storage_attestation","attest_author":"https://pith.science/pith/SZIRWIBGDCDKG6ONJHRIASHKA5/action/author_attestation","sign_citation":"https://pith.science/pith/SZIRWIBGDCDKG6ONJHRIASHKA5/action/citation_signature","submit_replication":"https://pith.science/pith/SZIRWIBGDCDKG6ONJHRIASHKA5/action/replication_record"}},"created_at":"2026-08-05T00:39:25.116328+00:00","updated_at":"2026-08-05T00:39:25.116328+00:00"}