{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:P3WII2RIRDMM2E23I6IOJXPS54","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":"95e6ae5ced3e7156f249893b3ea3a5660d173502854936614102cd22d1e7d013","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-14T15:50:01Z","title_canon_sha256":"084638641901ed903a190caf883afa608874d335be0d4c7f42c3e49e1b2745dc"},"schema_version":"1.0","source":{"id":"2605.15250","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.15250","created_at":"2026-05-20T00:00:48Z"},{"alias_kind":"arxiv_version","alias_value":"2605.15250v1","created_at":"2026-05-20T00:00:48Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.15250","created_at":"2026-05-20T00:00:48Z"},{"alias_kind":"pith_short_12","alias_value":"P3WII2RIRDMM","created_at":"2026-05-20T00:00:48Z"},{"alias_kind":"pith_short_16","alias_value":"P3WII2RIRDMM2E23","created_at":"2026-05-20T00:00:48Z"},{"alias_kind":"pith_short_8","alias_value":"P3WII2RI","created_at":"2026-05-20T00:00:48Z"}],"graph_snapshots":[{"event_id":"sha256:fb7912e65c773718b2bf96079b854ca7538f7c2258081dcb94fd421f93fa81d6","target":"graph","created_at":"2026-05-20T00:00:48Z","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":"A single set of GQLA weights pins the rooflines of both H100 (MQA-absorb, s_q=1) and H20 (GQA + MTP, s_q=2), while supporting up to 8-way zero-redundancy tensor parallelism on the GQA path, and compresses the per-token KV cache to 28.125% of the GQA baseline on the MQA-absorb path."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The two decoding paths remain algebraically equivalent after the TransGQLA conversion from a pretrained GQA checkpoint, so that accuracy and the claimed cache compression are preserved without any retraining or additional fine-tuning steps."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"GQLA exposes two algebraically equivalent decoding paths over one set of weights so a single model can hit roofline on both high-end and commodity GPUs while cutting KV cache size to 28% on the absorbed path."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Group-Query Latent Attention exposes two equivalent decoding paths from one set of weights for hardware-specific LLM inference."}],"snapshot_sha256":"0933b17f132a73117b521f516ceb760935bfc2dcdf995054757fcc61bee9dfa6"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[{"findings_count":0,"name":"doi_title_agreement","ran_at":"2026-05-19T17:31:18.462331Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"doi_compliance","ran_at":"2026-05-19T17:26:19.710935Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"claim_evidence","ran_at":"2026-05-19T15:01:54.172784Z","status":"completed","version":"1.0.0"},{"findings_count":0,"name":"ai_meta_artifact","ran_at":"2026-05-19T13:33:22.817469Z","status":"skipped","version":"1.0.0"}],"endpoint":"/pith/2605.15250/integrity.json","findings":[],"snapshot_sha256":"0556c384898048622e706d9fddd3da1f2d300e4ea0b608f6c1b2ae9410ff5045","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Multi-head Latent Attention (MLA), the attention used in DeepSeek-V2/V3, jointly compresses keys and values into a low-rank latent and matches the H100 roofline almost perfectly. Its trained weights, however, expose only one decoding path - an absorbed MQA form - which ties efficient inference to H100-class compute-bandwidth ratios, forfeits tensor parallelism along the head axis, and yields no Multi-Token Prediction (MTP) gain on commodity inference GPUs such as the export-restricted H20. We propose Group-Query Latent Attention (GQLA), a minimal modification of MLA whose trained weights expos","authors_text":"Fanxu Meng","cross_cats":["cs.AI"],"headline":"Group-Query Latent Attention exposes two equivalent decoding paths from one set of weights for hardware-specific LLM inference.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-14T15:50:01Z","title":"GQLA: Group-Query Latent Attention for Hardware-Adaptive Large Language Model Decoding"},"references":{"count":18,"internal_anchors":7,"resolved_work":18,"sample":[{"cited_arxiv_id":"1911.02150","doi":"","is_internal_anchor":true,"ref_index":1,"title":"Fast Transformer Decoding: One Write-Head is All You Need","work_id":"160ea164-b1d4-4adb-8ccb-a4655d8a0bb4","year":1911},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Proceedings of EMNLP , year=","work_id":"51df08b2-6fee-42c0-8f5d-c0b3c9766f25","year":null},{"cited_arxiv_id":"2405.04434","doi":"","is_internal_anchor":true,"ref_index":3,"title":"DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model","work_id":"1e1df141-cac8-47fd-b068-c4c96e51e331","year":null},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Advances in Neural Information Processing Systems , volume=","work_id":"547d76ec-c69c-4bfd-b1b2-c92f6be4b490","year":null},{"cited_arxiv_id":"2512.02556","doi":"","is_internal_anchor":true,"ref_index":5,"title":"DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models","work_id":"07c85cc5-4086-4abc-823b-6d0f4ff784d0","year":null}],"snapshot_sha256":"195fc96cd6f57a942634cae074d705a9664ba6b60756790531f7130e054e3dc8"},"source":{"id":"2605.15250","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-19T17:17:06.763524Z","id":"a4e98122-ca52-4802-b3b6-245c97842b2f","model_set":{"reader":"grok-4.3"},"one_line_summary":"GQLA exposes two algebraically equivalent decoding paths over one set of weights so a single model can hit roofline on both high-end and commodity GPUs while cutting KV cache size to 28% on the absorbed path.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Group-Query Latent Attention exposes two equivalent decoding paths from one set of weights for hardware-specific LLM inference.","strongest_claim":"A single set of GQLA weights pins the rooflines of both H100 (MQA-absorb, s_q=1) and H20 (GQA + MTP, s_q=2), while supporting up to 8-way zero-redundancy tensor parallelism on the GQA path, and compresses the per-token KV cache to 28.125% of the GQA baseline on the MQA-absorb path.","weakest_assumption":"The two decoding paths remain algebraically equivalent after the TransGQLA conversion from a pretrained GQA checkpoint, so that accuracy and the claimed cache compression are preserved without any retraining or additional fine-tuning steps."}},"verdict_id":"a4e98122-ca52-4802-b3b6-245c97842b2f"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:5c19a435936fe3bd45a793b5d0e5e562274c650de02aba6c4590cbe9c58d2b78","target":"record","created_at":"2026-05-20T00:00:48Z","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":"95e6ae5ced3e7156f249893b3ea3a5660d173502854936614102cd22d1e7d013","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2026-05-14T15:50:01Z","title_canon_sha256":"084638641901ed903a190caf883afa608874d335be0d4c7f42c3e49e1b2745dc"},"schema_version":"1.0","source":{"id":"2605.15250","kind":"arxiv","version":1}},"canonical_sha256":"7eec846a2888d8cd135b4790e4ddf2ef1a7cd54f3e4f2fccfb408dec59d6e21c","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"7eec846a2888d8cd135b4790e4ddf2ef1a7cd54f3e4f2fccfb408dec59d6e21c","first_computed_at":"2026-05-20T00:00:48.500751Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-20T00:00:48.500751Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"3k8Foom4MW+SaTancrmPq3eIPF7nvtlxajkF7Jbbmv3QXCFjNny9zY8HNDiCnoN4XO9chpY4JkaCiucQ3/ZPDA==","signature_status":"signed_v1","signed_at":"2026-05-20T00:00:48.501871Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.15250","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:5c19a435936fe3bd45a793b5d0e5e562274c650de02aba6c4590cbe9c58d2b78","sha256:fb7912e65c773718b2bf96079b854ca7538f7c2258081dcb94fd421f93fa81d6"],"state_sha256":"3362508d19355e93e573127fb654884c0442482fbf79f8ac3f757d4c7ae8d931"}