{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:5EBCKYJR64SMOIFCHSXDKEYHMR","short_pith_number":"pith:5EBCKYJR","schema_version":"1.0","canonical_sha256":"e902256131f724c720a23cae351307645a4d96898e4aed251630e1cd3a06bf1b","source":{"kind":"arxiv","id":"2402.05099","version":2},"attestation_state":"computed","paper":{"title":"Hydragen: High-Throughput LLM Inference with Shared Prefixes","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Azalia Mirhoseini, Bradley Brown, Christopher R\\'e, Daniel Y. Fu, Jordan Juravsky, Ryan Ehrlich","submitted_at":"2024-02-07T18:53:01Z","abstract_excerpt":"Transformer-based large language models (LLMs) are now deployed to hundreds of millions of users. LLM inference is commonly performed on batches of sequences that share a prefix, such as few-shot examples or a chatbot system prompt. Decoding in this large-batch setting can be bottlenecked by the attention operation, which reads large key-value (KV) caches from memory and computes inefficient matrix-vector products for every sequence in the batch. In this work, we introduce Hydragen, a hardware-aware exact implementation of attention with shared prefixes. Hydragen computes attention over the sh"},"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":"2402.05099","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2024-02-07T18:53:01Z","cross_cats_sorted":[],"title_canon_sha256":"190f585c1cd56f3371d17acc2a588f19f019e0f1d17ff5dffa4f1fde19a9a1da","abstract_canon_sha256":"133c1b84fcaf5eacf75464d9cea0f67f5afdd29eb70163758f6ad35470236570"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:18:05.053055Z","signature_b64":"WCG3HnF9Ct7NkWthbNUxRkfI325bgxCLrWGP95d8aYCvl+FL0DFLeGB25TMWY3wyAZ/cPwgaajkXblOSdIuGCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e902256131f724c720a23cae351307645a4d96898e4aed251630e1cd3a06bf1b","last_reissued_at":"2026-07-05T08:18:05.052571Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:18:05.052571Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Hydragen: High-Throughput LLM Inference with Shared Prefixes","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Azalia Mirhoseini, Bradley Brown, Christopher R\\'e, Daniel Y. Fu, Jordan Juravsky, Ryan Ehrlich","submitted_at":"2024-02-07T18:53:01Z","abstract_excerpt":"Transformer-based large language models (LLMs) are now deployed to hundreds of millions of users. LLM inference is commonly performed on batches of sequences that share a prefix, such as few-shot examples or a chatbot system prompt. Decoding in this large-batch setting can be bottlenecked by the attention operation, which reads large key-value (KV) caches from memory and computes inefficient matrix-vector products for every sequence in the batch. In this work, we introduce Hydragen, a hardware-aware exact implementation of attention with shared prefixes. Hydragen computes attention over the sh"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.05099","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/2402.05099/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":"2402.05099","created_at":"2026-07-05T08:18:05.052629+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.05099v2","created_at":"2026-07-05T08:18:05.052629+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.05099","created_at":"2026-07-05T08:18:05.052629+00:00"},{"alias_kind":"pith_short_12","alias_value":"5EBCKYJR64SM","created_at":"2026-07-05T08:18:05.052629+00:00"},{"alias_kind":"pith_short_16","alias_value":"5EBCKYJR64SMOIFC","created_at":"2026-07-05T08:18:05.052629+00:00"},{"alias_kind":"pith_short_8","alias_value":"5EBCKYJR","created_at":"2026-07-05T08:18:05.052629+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":12,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.23791","citing_title":"One Generator, Any Process: LLM-Conditioning for the LHC","ref_index":276,"is_internal_anchor":false},{"citing_arxiv_id":"2606.20537","citing_title":"Execution-State Capsules: Graph-Bound Execution-State Checkpoint and Restore for Low-Latency, Small-Batch, On-Device Physical-AI Serving","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2606.17059","citing_title":"Towards Distributed Inference of LLMs on a P2P Network","ref_index":5,"is_internal_anchor":false},{"citing_arxiv_id":"2606.23791","citing_title":"One Generator, Any Process: LLM-Conditioning for the LHC","ref_index":280,"is_internal_anchor":false},{"citing_arxiv_id":"2412.03594","citing_title":"BatchLLM: Optimizing Large Batched LLM Inference with Global Prefix Sharing and Throughput-oriented Token Batching","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2603.10726","citing_title":"PrefixWall: Mitigating Prefix Caching Side Channels in Shared LLM Systems","ref_index":38,"is_internal_anchor":false},{"citing_arxiv_id":"2605.02960","citing_title":"MoE-Prefill: Zero Redundancy Overheads in MoE Prefill Serving","ref_index":32,"is_internal_anchor":false},{"citing_arxiv_id":"2604.23477","citing_title":"SEMA-SQL: Beyond Traditional Relational Querying with Large Language Models","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2312.07104","citing_title":"SGLang: Efficient Execution of Structured Language Model Programs","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2407.21787","citing_title":"Large Language Monkeys: Scaling Inference Compute with Repeated Sampling","ref_index":37,"is_internal_anchor":false},{"citing_arxiv_id":"2604.23477","citing_title":"SEMA-SQL: Beyond Traditional Relational Querying with Large Language Models","ref_index":23,"is_internal_anchor":false},{"citing_arxiv_id":"2605.02960","citing_title":"MoE-Prefill: Zero Redundancy Overheads in MoE Prefill Serving","ref_index":32,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/5EBCKYJR64SMOIFCHSXDKEYHMR","json":"https://pith.science/pith/5EBCKYJR64SMOIFCHSXDKEYHMR.json","graph_json":"https://pith.science/api/pith-number/5EBCKYJR64SMOIFCHSXDKEYHMR/graph.json","events_json":"https://pith.science/api/pith-number/5EBCKYJR64SMOIFCHSXDKEYHMR/events.json","paper":"https://pith.science/paper/5EBCKYJR"},"agent_actions":{"view_html":"https://pith.science/pith/5EBCKYJR64SMOIFCHSXDKEYHMR","download_json":"https://pith.science/pith/5EBCKYJR64SMOIFCHSXDKEYHMR.json","view_paper":"https://pith.science/paper/5EBCKYJR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.05099&json=true","fetch_graph":"https://pith.science/api/pith-number/5EBCKYJR64SMOIFCHSXDKEYHMR/graph.json","fetch_events":"https://pith.science/api/pith-number/5EBCKYJR64SMOIFCHSXDKEYHMR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/5EBCKYJR64SMOIFCHSXDKEYHMR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/5EBCKYJR64SMOIFCHSXDKEYHMR/action/storage_attestation","attest_author":"https://pith.science/pith/5EBCKYJR64SMOIFCHSXDKEYHMR/action/author_attestation","sign_citation":"https://pith.science/pith/5EBCKYJR64SMOIFCHSXDKEYHMR/action/citation_signature","submit_replication":"https://pith.science/pith/5EBCKYJR64SMOIFCHSXDKEYHMR/action/replication_record"}},"created_at":"2026-07-05T08:18:05.052629+00:00","updated_at":"2026-07-05T08:18:05.052629+00:00"}