{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2025:3R3KCDTMVQK333KEAAEZTZBUHW","short_pith_number":"pith:3R3KCDTM","canonical_record":{"source":{"id":"2506.09342","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-06-11T02:48:16Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"5a074719724464aa7c9dc1b2f40bd060e826de30d19db5f3fa4f0a85acc810e6","abstract_canon_sha256":"4ab9bf96263d862d3c8386db34e887b54f46481f64edcd24d6fda7883377297a"},"schema_version":"1.0"},"canonical_sha256":"dc76a10e6cac15bded44000999e4343dae6f3f911cdcee3b2bfe6fbca7335d8f","source":{"kind":"arxiv","id":"2506.09342","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2506.09342","created_at":"2026-07-05T11:21:49Z"},{"alias_kind":"arxiv_version","alias_value":"2506.09342v2","created_at":"2026-07-05T11:21:49Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.09342","created_at":"2026-07-05T11:21:49Z"},{"alias_kind":"pith_short_12","alias_value":"3R3KCDTMVQK3","created_at":"2026-07-05T11:21:49Z"},{"alias_kind":"pith_short_16","alias_value":"3R3KCDTMVQK333KE","created_at":"2026-07-05T11:21:49Z"},{"alias_kind":"pith_short_8","alias_value":"3R3KCDTM","created_at":"2026-07-05T11:21:49Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2025:3R3KCDTMVQK333KEAAEZTZBUHW","target":"record","payload":{"canonical_record":{"source":{"id":"2506.09342","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-06-11T02:48:16Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"5a074719724464aa7c9dc1b2f40bd060e826de30d19db5f3fa4f0a85acc810e6","abstract_canon_sha256":"4ab9bf96263d862d3c8386db34e887b54f46481f64edcd24d6fda7883377297a"},"schema_version":"1.0"},"canonical_sha256":"dc76a10e6cac15bded44000999e4343dae6f3f911cdcee3b2bfe6fbca7335d8f","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:21:49.313911Z","signature_b64":"/69NxrWabY0iCb78XR7lzlePdSt5iHfgivJCT9SWdjazxTYoJSXFrro8M+6gBQz7zZLjAoSmSH7UOgeGPg6/BQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"dc76a10e6cac15bded44000999e4343dae6f3f911cdcee3b2bfe6fbca7335d8f","last_reissued_at":"2026-07-05T11:21:49.313337Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:21:49.313337Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2506.09342","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T11:21:49Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"PhhROqx9biehbNsC4fHDXxrM6ppO4VVxLKXdgsO+hO1fsf11OQ/Li9ox0j1KU1yYrh9icxuecVZPnnK018O1Cg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-07T11:43:09.839524Z"},"content_sha256":"55bb5ca68c0d2a3e66a8c154e4fb29a9390cdac206252228a257875d1848b111","schema_version":"1.0","event_id":"sha256:55bb5ca68c0d2a3e66a8c154e4fb29a9390cdac206252228a257875d1848b111"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2025:3R3KCDTMVQK333KEAAEZTZBUHW","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Latent Multi-Head Attention for Small Language Models","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Rajat Dandekar, Raj Dandekar, Sreedath Panat, Sushant Mehta","submitted_at":"2025-06-11T02:48:16Z","abstract_excerpt":"We present the first comprehensive study of latent multi-head attention (MLA) for small language models, revealing interesting efficiency-quality trade-offs. Training 30M-parameter GPT models on 100,000 synthetic stories, we benchmark three architectural variants: standard multi-head attention (MHA), MLA, and MLA with rotary positional embeddings (MLA+RoPE). Our key finding is that MLA+RoPE with half-rank latent dimensions (r = d/2) achieves a 45% KV-cache memory reduction while incurring only a 0.3% increase in validation loss (essentially matching MHA quality)- a Pareto improvement for memor"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.09342","kind":"arxiv","version":2},"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/2506.09342/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"},"verdict_id":null},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T11:21:49Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"2aJejMUAs5+15ZW6y7qetR8pRBjX8RRCozZEOXujbqUaT/5unqeRkN5syofeKIC2fs8nl+/WUeHvFXrP079+DQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-07T11:43:09.839905Z"},"content_sha256":"2d549df02ab26a9bac2a556c893045854d601dcc85a1371b4cc5698540f59287","schema_version":"1.0","event_id":"sha256:2d549df02ab26a9bac2a556c893045854d601dcc85a1371b4cc5698540f59287"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/3R3KCDTMVQK333KEAAEZTZBUHW/bundle.json","state_url":"https://pith.science/pith/3R3KCDTMVQK333KEAAEZTZBUHW/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/3R3KCDTMVQK333KEAAEZTZBUHW/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-07-07T11:43:09Z","links":{"resolver":"https://pith.science/pith/3R3KCDTMVQK333KEAAEZTZBUHW","bundle":"https://pith.science/pith/3R3KCDTMVQK333KEAAEZTZBUHW/bundle.json","state":"https://pith.science/pith/3R3KCDTMVQK333KEAAEZTZBUHW/state.json","well_known_bundle":"https://pith.science/.well-known/pith/3R3KCDTMVQK333KEAAEZTZBUHW/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:3R3KCDTMVQK333KEAAEZTZBUHW","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":"4ab9bf96263d862d3c8386db34e887b54f46481f64edcd24d6fda7883377297a","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-06-11T02:48:16Z","title_canon_sha256":"5a074719724464aa7c9dc1b2f40bd060e826de30d19db5f3fa4f0a85acc810e6"},"schema_version":"1.0","source":{"id":"2506.09342","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2506.09342","created_at":"2026-07-05T11:21:49Z"},{"alias_kind":"arxiv_version","alias_value":"2506.09342v2","created_at":"2026-07-05T11:21:49Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.09342","created_at":"2026-07-05T11:21:49Z"},{"alias_kind":"pith_short_12","alias_value":"3R3KCDTMVQK3","created_at":"2026-07-05T11:21:49Z"},{"alias_kind":"pith_short_16","alias_value":"3R3KCDTMVQK333KE","created_at":"2026-07-05T11:21:49Z"},{"alias_kind":"pith_short_8","alias_value":"3R3KCDTM","created_at":"2026-07-05T11:21:49Z"}],"graph_snapshots":[{"event_id":"sha256:2d549df02ab26a9bac2a556c893045854d601dcc85a1371b4cc5698540f59287","target":"graph","created_at":"2026-07-05T11:21:49Z","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":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2506.09342/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"We present the first comprehensive study of latent multi-head attention (MLA) for small language models, revealing interesting efficiency-quality trade-offs. Training 30M-parameter GPT models on 100,000 synthetic stories, we benchmark three architectural variants: standard multi-head attention (MHA), MLA, and MLA with rotary positional embeddings (MLA+RoPE). Our key finding is that MLA+RoPE with half-rank latent dimensions (r = d/2) achieves a 45% KV-cache memory reduction while incurring only a 0.3% increase in validation loss (essentially matching MHA quality)- a Pareto improvement for memor","authors_text":"Rajat Dandekar, Raj Dandekar, Sreedath Panat, Sushant Mehta","cross_cats":["cs.AI"],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-06-11T02:48:16Z","title":"Latent Multi-Head Attention for Small Language Models"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.09342","kind":"arxiv","version":2},"verdict":{"created_at":null,"id":null,"model_set":{},"one_line_summary":"","pipeline_version":null,"pith_extraction_headline":"","strongest_claim":"","weakest_assumption":""}},"verdict_id":null}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:55bb5ca68c0d2a3e66a8c154e4fb29a9390cdac206252228a257875d1848b111","target":"record","created_at":"2026-07-05T11:21:49Z","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":"4ab9bf96263d862d3c8386db34e887b54f46481f64edcd24d6fda7883377297a","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-06-11T02:48:16Z","title_canon_sha256":"5a074719724464aa7c9dc1b2f40bd060e826de30d19db5f3fa4f0a85acc810e6"},"schema_version":"1.0","source":{"id":"2506.09342","kind":"arxiv","version":2}},"canonical_sha256":"dc76a10e6cac15bded44000999e4343dae6f3f911cdcee3b2bfe6fbca7335d8f","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"dc76a10e6cac15bded44000999e4343dae6f3f911cdcee3b2bfe6fbca7335d8f","first_computed_at":"2026-07-05T11:21:49.313337Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T11:21:49.313337Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"/69NxrWabY0iCb78XR7lzlePdSt5iHfgivJCT9SWdjazxTYoJSXFrro8M+6gBQz7zZLjAoSmSH7UOgeGPg6/BQ==","signature_status":"signed_v1","signed_at":"2026-07-05T11:21:49.313911Z","signed_message":"canonical_sha256_bytes"},"source_id":"2506.09342","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:55bb5ca68c0d2a3e66a8c154e4fb29a9390cdac206252228a257875d1848b111","sha256:2d549df02ab26a9bac2a556c893045854d601dcc85a1371b4cc5698540f59287"],"state_sha256":"3fe79d80e940ba9844f435205f42c0bbcec971c9bf05d0be2b754903c1d68416"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"ZM1w6TtgF/WKE2EJb7ON30pbKVqd301k8bpXn2VOa0TH15+kVpIbTknQFXooctuXsq4XHdwcjm6jiqzOxWg1CQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-07-07T11:43:09.841899Z","bundle_sha256":"7c58db2c0644bba5cdb7233cc4276789f5176b2ee1262e5ff3d38d04cedcb675"}}