{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/optimizing-model-selection-for-compound-ai","title":"Optimizing Model Selection for Compound AI Systems","arxiv_id":"2502.14815","date":"2025-02-20","proceeding":null,"authors":["Lingjiao Chen","Jared Quincy Davis","Boris Hanin","Peter Bailis","Matei Zaharia","James Zou","Ion Stoica"],"abstract":"Compound AI systems that combine multiple LLM calls, such as self-refine and multi-agent-debate, achieve strong performance on many AI tasks. We address a core question in optimizing compound systems: for each LLM call or module in the system, how should one decide which LLM to use? We show that these LLM choices have a large effect on quality, but the search space is exponential. We propose LLMSelector, an efficient framework for model selection in compound systems, which leverages two key empirical insights: (i) end-to-end performance is often monotonic in how well each module performs, with all other modules held fixed, and (ii) per-module performance can be estimated accurately by an LLM. Building upon these insights, LLMSelector iteratively selects one module and allocates to it the model with the highest module-wise performance, as estimated by an LLM, until no further gain is possible. LLMSelector is applicable to any compound system with a bounded number of modules, and its number of API calls scales linearly with the number of modules, achieving high-quality model allocation both empirically and theoretically. Experiments with popular compound systems such as multi-agent debate and self-refine using LLMs such as GPT-4o, Claude 3.5 Sonnet and Gemini 1.5 show that LLMSelector confers 5%-70% accuracy gains compared to using the same LLM for all modules.","url_abs":"https://arxiv.org/abs/2502.14815v1","url_pdf":"https://arxiv.org/pdf/2502.14815v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"optimizing-model-selection-for-compound-ai","repo_url":"https://github.com/LLMSELECTOR/LLMSELECTOR","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"none","reach":{"status":"ok","spdx":"Apache-2.0"}}],"tasks":[{"task_slug":"model-selection","task_name":"Model Selection"},{"task_slug":"model","task_name":"model"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2502.14815","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2502.14815"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-24T18:15:14+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/LLMSELECTOR/LLMSELECTOR","reach":{"status":"ok","spdx":"Apache-2.0"}}],"summary":{"ran_draft_wrong":1,"unverified":9},"by_repo_kind":{"official":{"samples":10,"ran":1,"repositories":1}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted","community":"Not in the archive's code links for this paper; a community repository Syntology harvested"},"n_pointer_only_for_licence":0,"samples":[{"code_sha256_prefix":"0892f960b112ef52","entry":"strip_latex","repo":"LLMSELECTOR/LLMSELECTOR","repo_kind":"official","path":"llmselector/llmselector/compoundai/metric.py","file_url":"https://github.com/LLMSELECTOR/LLMSELECTOR/blob/HEAD/llmselector/llmselector/compoundai/metric.py","link_basis":"harvester_set","language":"python","status":"ran_draft_wrong","verification_level":1,"contract_check":"OUTPUT_MISDECLARED","metamorphic_tier":"deterministic","behaviour_fingerprint":true,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"0892f960b112ef52"}},{"code_sha256_prefix":"96362ca151f15497","entry":"Get_Generate","repo":"LLMSELECTOR/LLMSELECTOR","repo_kind":"official","path":"llmselector/llmselector/compoundai/llm.py","file_url":"https://github.com/LLMSELECTOR/LLMSELECTOR/blob/HEAD/llmselector/llmselector/compoundai/llm.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"96362ca151f15497"}},{"code_sha256_prefix":"075a2a4962fab858","entry":"compute_score","repo":"LLMSELECTOR/LLMSELECTOR","repo_kind":"official","path":"llmselector/llmselector/compoundai/metric.py","file_url":"https://github.com/LLMSELECTOR/LLMSELECTOR/blob/HEAD/llmselector/llmselector/compoundai/metric.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"075a2a4962fab858"}},{"code_sha256_prefix":"1a27128efa695483","entry":"compute_tag","repo":"LLMSELECTOR/LLMSELECTOR","repo_kind":"official","path":"llmselector/llmselector/compoundai/metric.py","file_url":"https://github.com/LLMSELECTOR/LLMSELECTOR/blob/HEAD/llmselector/llmselector/compoundai/metric.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"1a27128efa695483"}},{"code_sha256_prefix":"ae7782a357cf3a90","entry":"extract_ans","repo":"LLMSELECTOR/LLMSELECTOR","repo_kind":"official","path":"llmselector/llmselector/compoundai/module/debate.py","file_url":"https://github.com/LLMSELECTOR/LLMSELECTOR/blob/HEAD/llmselector/llmselector/compoundai/module/debate.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"ae7782a357cf3a90"}},{"code_sha256_prefix":"4ec0ffdf43c1c1c5","entry":"extract_ans","repo":"LLMSELECTOR/LLMSELECTOR","repo_kind":"official","path":"llmselector/llmselector/compoundai/module/majorityvote.py","file_url":"https://github.com/LLMSELECTOR/LLMSELECTOR/blob/HEAD/llmselector/llmselector/compoundai/module/majorityvote.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"4ec0ffdf43c1c1c5"}},{"code_sha256_prefix":"5672cff2de3b190b","entry":"extract_final_answer","repo":"LLMSELECTOR/LLMSELECTOR","repo_kind":"official","path":"llmselector/llmselector/compoundai/diagnoser.py","file_url":"https://github.com/LLMSELECTOR/LLMSELECTOR/blob/HEAD/llmselector/llmselector/compoundai/diagnoser.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"5672cff2de3b190b"}},{"code_sha256_prefix":"05b617dbf3b25a2a","entry":"my_decode","repo":"LLMSELECTOR/LLMSELECTOR","repo_kind":"official","path":"llmselector/llmselector/compoundai/llm.py","file_url":"https://github.com/LLMSELECTOR/LLMSELECTOR/blob/HEAD/llmselector/llmselector/compoundai/llm.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"05b617dbf3b25a2a"}},{"code_sha256_prefix":"f2fc23dea676ed77","entry":"my_encode","repo":"LLMSELECTOR/LLMSELECTOR","repo_kind":"official","path":"llmselector/llmselector/compoundai/llm.py","file_url":"https://github.com/LLMSELECTOR/LLMSELECTOR/blob/HEAD/llmselector/llmselector/compoundai/llm.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"f2fc23dea676ed77"}},{"code_sha256_prefix":"d957a2e9f35bc763","entry":"resize_to_max_pixels","repo":"LLMSELECTOR/LLMSELECTOR","repo_kind":"official","path":"llmselector/llmselector/compoundai/module/directgen.py","file_url":"https://github.com/LLMSELECTOR/LLMSELECTOR/blob/HEAD/llmselector/llmselector/compoundai/module/directgen.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"d957a2e9f35bc763"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}