{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2025:N3HS76Y2IWUSH3K6Y47ZQQ7ICN","short_pith_number":"pith:N3HS76Y2","canonical_record":{"source":{"id":"2502.07864","kind":"arxiv","version":5},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-02-11T18:20:18Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"78b9c6bf6c293e52e8097feffe0782d14a535ec9040f99c17d61f710dea7351e","abstract_canon_sha256":"05481715f94cc16c04b85adb37f3435e67980c1708aea6839fc6cf8ed6aad9e9"},"schema_version":"1.0"},"canonical_sha256":"6ecf2ffb1a45a923ed5ec73f9843e8135dadb1946080a84463ed33a67b1a7d52","source":{"kind":"arxiv","id":"2502.07864","version":5},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2502.07864","created_at":"2026-07-05T11:20:09Z"},{"alias_kind":"arxiv_version","alias_value":"2502.07864v5","created_at":"2026-07-05T11:20:09Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.07864","created_at":"2026-07-05T11:20:09Z"},{"alias_kind":"pith_short_12","alias_value":"N3HS76Y2IWUS","created_at":"2026-07-05T11:20:09Z"},{"alias_kind":"pith_short_16","alias_value":"N3HS76Y2IWUSH3K6","created_at":"2026-07-05T11:20:09Z"},{"alias_kind":"pith_short_8","alias_value":"N3HS76Y2","created_at":"2026-07-05T11:20:09Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2025:N3HS76Y2IWUSH3K6Y47ZQQ7ICN","target":"record","payload":{"canonical_record":{"source":{"id":"2502.07864","kind":"arxiv","version":5},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-02-11T18:20:18Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"78b9c6bf6c293e52e8097feffe0782d14a535ec9040f99c17d61f710dea7351e","abstract_canon_sha256":"05481715f94cc16c04b85adb37f3435e67980c1708aea6839fc6cf8ed6aad9e9"},"schema_version":"1.0"},"canonical_sha256":"6ecf2ffb1a45a923ed5ec73f9843e8135dadb1946080a84463ed33a67b1a7d52","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:20:09.074157Z","signature_b64":"Q8dANKOTqJkfoF5zCfiDLoj7SPvVATSWvTr0ngHfondaYqB8e1PImKuuZ147WCiH7P4htl083BKKQth2C5sSCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"6ecf2ffb1a45a923ed5ec73f9843e8135dadb1946080a84463ed33a67b1a7d52","last_reissued_at":"2026-07-05T11:20:09.073499Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:20:09.073499Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2502.07864","source_version":5,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T11:20:09Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"o7ScAv71xfGrYSe7mnzfl7y0CMol78bO02aKZKdOy7/DVXDqnvvtODJEHnNRZm46XPr1l194sMVXQ48iwFyVBw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-05T12:59:15.887936Z"},"content_sha256":"3048ffae91ea1a3406904fe6c1c9b79f9ffead5622eae5689fa59e6b13422090","schema_version":"1.0","event_id":"sha256:3048ffae91ea1a3406904fe6c1c9b79f9ffead5622eae5689fa59e6b13422090"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2025:N3HS76Y2IWUSH3K6Y47ZQQ7ICN","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"TransMLA: Multi-Head Latent Attention Is All You Need","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"Fanxu Meng, Muhan Zhang, Pingzhi Tang, Xiaojuan Tang, Xing Sun, Zengwei Yao","submitted_at":"2025-02-11T18:20:18Z","abstract_excerpt":"In this paper, we present TransMLA, a framework that seamlessly converts any GQA-based pre-trained model into an MLA-based model. Our approach enables direct compatibility with DeepSeek's codebase, allowing these models to fully leverage DeepSeek-specific optimizations such as vLLM and SGlang. By compressing 93% of the KV cache in LLaMA-2-7B, TransMLA achieves a 10.6x inference speedup at an 8K context length while preserving meaningful output quality. Additionally, the model requires only 6 billion tokens for fine-tuning to regain performance on par with the original across multiple benchmark"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.07864","kind":"arxiv","version":5},"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/2502.07864/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":"2026-07-05T11:20:09Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"RzqcnSsiBiMegghscUM22XVZ8m+UDmJ2P71RYij80FpB3ufQNUImRqni8CmiZsv111GhsUuOjt4esbHFjo8MAw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-05T12:59:15.888652Z"},"content_sha256":"3dd2828a5c3bbd00837ae6ad193a569ddfe58c3d3daa3322589f780ed4690b7f","schema_version":"1.0","event_id":"sha256:3dd2828a5c3bbd00837ae6ad193a569ddfe58c3d3daa3322589f780ed4690b7f"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/N3HS76Y2IWUSH3K6Y47ZQQ7ICN/bundle.json","state_url":"https://pith.science/pith/N3HS76Y2IWUSH3K6Y47ZQQ7ICN/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/N3HS76Y2IWUSH3K6Y47ZQQ7ICN/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":"2026-08-05T12:59:15Z","links":{"resolver":"https://pith.science/pith/N3HS76Y2IWUSH3K6Y47ZQQ7ICN","bundle":"https://pith.science/pith/N3HS76Y2IWUSH3K6Y47ZQQ7ICN/bundle.json","state":"https://pith.science/pith/N3HS76Y2IWUSH3K6Y47ZQQ7ICN/state.json","well_known_bundle":"https://pith.science/.well-known/pith/N3HS76Y2IWUSH3K6Y47ZQQ7ICN/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2025:N3HS76Y2IWUSH3K6Y47ZQQ7ICN","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":"05481715f94cc16c04b85adb37f3435e67980c1708aea6839fc6cf8ed6aad9e9","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-02-11T18:20:18Z","title_canon_sha256":"78b9c6bf6c293e52e8097feffe0782d14a535ec9040f99c17d61f710dea7351e"},"schema_version":"1.0","source":{"id":"2502.07864","kind":"arxiv","version":5}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2502.07864","created_at":"2026-07-05T11:20:09Z"},{"alias_kind":"arxiv_version","alias_value":"2502.07864v5","created_at":"2026-07-05T11:20:09Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.07864","created_at":"2026-07-05T11:20:09Z"},{"alias_kind":"pith_short_12","alias_value":"N3HS76Y2IWUS","created_at":"2026-07-05T11:20:09Z"},{"alias_kind":"pith_short_16","alias_value":"N3HS76Y2IWUSH3K6","created_at":"2026-07-05T11:20:09Z"},{"alias_kind":"pith_short_8","alias_value":"N3HS76Y2","created_at":"2026-07-05T11:20:09Z"}],"graph_snapshots":[{"event_id":"sha256:3dd2828a5c3bbd00837ae6ad193a569ddfe58c3d3daa3322589f780ed4690b7f","target":"graph","created_at":"2026-07-05T11:20:09Z","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/2502.07864/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"In this paper, we present TransMLA, a framework that seamlessly converts any GQA-based pre-trained model into an MLA-based model. Our approach enables direct compatibility with DeepSeek's codebase, allowing these models to fully leverage DeepSeek-specific optimizations such as vLLM and SGlang. By compressing 93% of the KV cache in LLaMA-2-7B, TransMLA achieves a 10.6x inference speedup at an 8K context length while preserving meaningful output quality. Additionally, the model requires only 6 billion tokens for fine-tuning to regain performance on par with the original across multiple benchmark","authors_text":"Fanxu Meng, Muhan Zhang, Pingzhi Tang, Xiaojuan Tang, Xing Sun, Zengwei Yao","cross_cats":["cs.AI"],"headline":"","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-02-11T18:20:18Z","title":"TransMLA: Multi-Head Latent Attention Is All You Need"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.07864","kind":"arxiv","version":5},"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":"sha256:3048ffae91ea1a3406904fe6c1c9b79f9ffead5622eae5689fa59e6b13422090","target":"record","created_at":"2026-07-05T11:20:09Z","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":"05481715f94cc16c04b85adb37f3435e67980c1708aea6839fc6cf8ed6aad9e9","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2025-02-11T18:20:18Z","title_canon_sha256":"78b9c6bf6c293e52e8097feffe0782d14a535ec9040f99c17d61f710dea7351e"},"schema_version":"1.0","source":{"id":"2502.07864","kind":"arxiv","version":5}},"canonical_sha256":"6ecf2ffb1a45a923ed5ec73f9843e8135dadb1946080a84463ed33a67b1a7d52","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"6ecf2ffb1a45a923ed5ec73f9843e8135dadb1946080a84463ed33a67b1a7d52","first_computed_at":"2026-07-05T11:20:09.073499Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T11:20:09.073499Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"Q8dANKOTqJkfoF5zCfiDLoj7SPvVATSWvTr0ngHfondaYqB8e1PImKuuZ147WCiH7P4htl083BKKQth2C5sSCA==","signature_status":"signed_v1","signed_at":"2026-07-05T11:20:09.074157Z","signed_message":"canonical_sha256_bytes"},"source_id":"2502.07864","source_kind":"arxiv","source_version":5}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:3048ffae91ea1a3406904fe6c1c9b79f9ffead5622eae5689fa59e6b13422090","sha256:3dd2828a5c3bbd00837ae6ad193a569ddfe58c3d3daa3322589f780ed4690b7f"],"state_sha256":"84a0ca170ab2cab3679ee50d2038622a7c8b6503b82e5226355ea52a6e741c8d"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"A7D6E9Z5TadHEQG9GKT9bMFNterDzgghdsjR1J2UPhTMqAIuAkHglSsnoF0VFatRDUqecP9ATa7IPxxqHvG2Aw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-05T12:59:15.893436Z","bundle_sha256":"dda705b2a5c6a998fe5a366de52ea496fcc47286cd84bb5fb3f9e1f76dbdaaea"}}