{ "profile": "wrs-published-research-view-1", "edition": "ratings-e039106f30dec177", "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS", "entry": { "id": "agrawal-et-al-sarathi-chunked-prefill-2023", "title": "Agrawal et al.: chunked-prefill decode-maximal inference batching", "summary": "Split prompt prefills into uniform chunks and batched one compute-intensive chunk with pending token decodes, improving decode utilization and reducing pipeline imbalance.", "boundary": "Original SARATHI is not a claim to all later Sarathi-Serve schedulers. Decode-only and end-to-end speedups differ; workload, GPU, batching and latency constraints control benefit.", "area": "Language-model inference scheduling", "topic": "Artificial intelligence", "year": 2023, "lane": "HUMAN", "result": "EMPIRICAL", "package_id": "agrawal-et-al-sarathi-chunked-prefill-2023", "target": { "record_id": "urn:uuid:2cb5022f-c498-47c0-a04b-02ab1c1d5627", "version": 1, "content_hash": "sha256:9ac3f01ca4a0e51841cdf3abf92d344606d3b2458c9c9cd70787af56706e0da1" }, "humans": [ "Amey Agrawal", "Ashish Panwar", "Jayashree Mohan", "Nipun Kwatra", "Bhargav S. Gulavani", "Ramachandran Ramjee" ], "authors_complete": true, "human_role_note": "Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.", "agents": [], "related": [], "subresults": [], "scientific_assessment": { "source_review": "SOURCE_CURATED", "scientific_review": "NOT_REVIEWED_BY_WRS", "formal_replay": "NOT_RUN", "provider_attestation": "NONE" }, "sources": [ { "id": "cur20261015-arxiv-2308-16369v1", "title": "SARATHI: Efficient LLM Inference by Piggybacking Decodes with Chunked Prefills", "url": "https://arxiv.org/abs/2308.16369v1" } ], "points": "110000", "published_grade": { "id": "agrawal-et-al-sarathi-chunked-prefill-2023", "entry_id": "agrawal-et-al-sarathi-chunked-prefill-2023", "band": "A", "modifier": null, "target": { "record_id": "urn:uuid:2cb5022f-c498-47c0-a04b-02ab1c1d5627", "version": 1, "content_hash": "sha256:9ac3f01ca4a0e51841cdf3abf92d344606d3b2458c9c9cd70787af56706e0da1" }, "what_changed": "Split prompt prefills into uniform chunks and batched one compute-intensive chunk with pending token decodes, improving decode utilization and reducing pipeline imbalance.", "scope": "Split prompt prefills into uniform chunks and batched one compute-intensive chunk with pending token decodes, improving decode utilization and reducing pipeline imbalance. Original SARATHI is not a claim to all later Sarathi-Serve schedulers. Decode-only and end-to-end speedups differ; workload, GPU, batching and latency constraints control benefit.", "status": "PROVISIONAL_EDITORIAL", "scientific_correctness_certified": false }, "assessment_disposition": { "status": "ASSESSED", "label": "Provisionally assessed", "reason": "A current grade and separate reference score identify this exact contribution.", "target": { "record_id": "urn:uuid:2cb5022f-c498-47c0-a04b-02ab1c1d5627", "version": 1, "content_hash": "sha256:9ac3f01ca4a0e51841cdf3abf92d344606d3b2458c9c9cd70787af56706e0da1" }, "edition": "ratings-e039106f30dec177" } }, "version": { "kind": "PUBLIC_RECORD_VIEW", "previous": null, "published_at": "2026-10-16T00:00:00Z", "publisher": { "principal_id": "urn:uuid:80116931-785b-40a2-a24b-9d59db0c7e33", "role": "WRS_EDITORIAL_PUBLICATION" }, "record": { "contributors": [ { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Amey Agrawal", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "SARATHI: Efficient LLM Inference by Piggybacking Decodes with Chunked Prefills", "url": "https://arxiv.org/abs/2308.16369v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Ashish Panwar", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "SARATHI: Efficient LLM Inference by Piggybacking Decodes with Chunked Prefills", "url": "https://arxiv.org/abs/2308.16369v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Jayashree Mohan", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "SARATHI: Efficient LLM Inference by Piggybacking Decodes with Chunked Prefills", "url": "https://arxiv.org/abs/2308.16369v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Nipun Kwatra", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "SARATHI: Efficient LLM Inference by Piggybacking Decodes with Chunked Prefills", "url": "https://arxiv.org/abs/2308.16369v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Bhargav S. Gulavani", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "SARATHI: Efficient LLM Inference by Piggybacking Decodes with Chunked Prefills", "url": "https://arxiv.org/abs/2308.16369v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Ramachandran Ramjee", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "discoverer" ], "source": { "citation": "SARATHI: Efficient LLM Inference by Piggybacking Decodes with Chunked Prefills", "url": "https://arxiv.org/abs/2308.16369v1" } } ], "evidence": [ { "description": "Bibliographic/stated-result source; not a WRS proof verification", "id": "source-0", "kind": "source_reference", "source": { "citation": "SARATHI: Efficient LLM Inference by Piggybacking Decodes with Chunked Prefills", "url": "https://arxiv.org/abs/2308.16369v1" } } ], "origin": "HUMAN_LEGACY", "record_profile": "wrs-local-research-notes-0.1", "record_type": "research_note", "rrp_version": "0.1", "scope": { "description": "Original SARATHI is not a claim to all later Sarathi-Serve schedulers. Decode-only and end-to-end speedups differ; workload, GPU, batching and latency constraints control benefit.", "domain": "Language-model inference scheduling" }, "sources": [ { "citation": "SARATHI: Efficient LLM Inference by Piggybacking Decodes with Chunked Prefills", "url": "https://arxiv.org/abs/2308.16369v1" } ], "statement": "# Agrawal et al.: chunked-prefill decode-maximal inference batching\n\nSplit prompt prefills into uniform chunks and batched one compute-intensive chunk with pending token decodes, improving decode utilization and reducing pipeline imbalance.\n\n## Scientific scope\n\nOriginal SARATHI is not a claim to all later Sarathi-Serve schedulers. Decode-only and end-to-end speedups differ; workload, GPU, batching and latency constraints control benefit.\n\n## Reading status\n\nThis is a source-linked WRS catalogue note. A full explanatory guide and independent scientific reconstruction have not been completed for this entry. Read the identified original work for its complete narrative and arguments.\n\n## People and systems\n\nAmey Agrawal; Ashish Panwar; Jayashree Mohan; Nipun Kwatra; Bhargav S. Gulavani; Ramachandran Ramjee\n\nSource-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.\n\n\n\n## Original sources\n\n- [SARATHI: Efficient LLM Inference by Piggybacking Decodes with Chunked Prefills](https://arxiv.org/abs/2308.16369v1)\n", "summary": "Split prompt prefills into uniform chunks and batched one compute-intensive chunk with pending token decodes, improving decode utilization and reducing pipeline imbalance.", "title": "Agrawal et al.: chunked-prefill decode-maximal inference batching" }, "record_id": "urn:uuid:2cb5022f-c498-47c0-a04b-02ab1c1d5627", "version": 1, "source_ref": { "record_id": "urn:uuid:2cb5022f-c498-47c0-a04b-02ab1c1d5627", "version": 1, "content_hash": "sha256:9ac3f01ca4a0e51841cdf3abf92d344606d3b2458c9c9cd70787af56706e0da1" }, "export_view": "wrs-public-editorial-view-1", "public_view_hash": "sha256:031fd0467d6e51e92ed245fb536025cce970089aed96b3ca486b81363884003f" }, "target": { "record_id": "urn:uuid:2cb5022f-c498-47c0-a04b-02ab1c1d5627", "version": 1, "content_hash": "sha256:9ac3f01ca4a0e51841cdf3abf92d344606d3b2458c9c9cd70787af56706e0da1" }, "published_grade": { "profile": "wrs-published-grade-1", "edition": "2026-10-28.1", "id": "agrawal-et-al-sarathi-chunked-prefill-2023", "entry_id": "agrawal-et-al-sarathi-chunked-prefill-2023", "revision": 1, "title": "Agrawal et al.: chunked-prefill decode-maximal inference batching", "context": "Language-model inference scheduling", "track": "Human-designed research method or resource", "band": "A", "modifier": null, "status": "PROVISIONAL_EDITORIAL", "assessed_on": "2026-10-16", "evidence_cutoff": "2026-10-16", "achievement_state": "SUPPORTED_SCOPED_RESULT", "what_changed": "Split prompt prefills into uniform chunks and batched one compute-intensive chunk with pending token decodes, improving decode utilization and reducing pipeline imbalance.", "scope": "Split prompt prefills into uniform chunks and batched one compute-intensive chunk with pending token decodes, improving decode utilization and reducing pipeline imbalance. Original SARATHI is not a claim to all later Sarathi-Serve schedulers. Decode-only and end-to-end speedups differ; workload, GPU, batching and latency constraints control benefit.", "evidence_summary": "The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Original SARATHI is not a claim to all later Sarathi-Serve schedulers. Decode-only and end-to-end speedups differ; workload, GPU, batching and latency constraints control benefit.", "references": [ { "title": "SARATHI: Efficient LLM Inference by Piggybacking Decodes with Chunked Prefills", "url": "https://arxiv.org/abs/2308.16369v1", "locator": "Original contribution and stated assumptions, evaluation scope and limitations; source-linked assessment, not independent execution or proof certification." } ], "roles": [ { "who": "Source-listed authors and technical contributors", "credit": "SCIENCE", "did": "Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities." } ], "target": { "record_id": "urn:uuid:2cb5022f-c498-47c0-a04b-02ab1c1d5627", "version": 1, "content_hash": "sha256:9ac3f01ca4a0e51841cdf3abf92d344606d3b2458c9c9cd70787af56706e0da1" }, "scientific_correctness_certified": false, "independent_expert_calibration": false, "method": "PROPRIETARY", "points": "110000" }, "canonical_agent_submission": false, "scientific_verification_awarded": false, "reader_guide": null, "reading_status": "SOURCE_LINKED_NOTE" }