{"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/designing-effective-sparse-expert-models","title":"ST-MoE: Designing Stable and Transferable Sparse Expert Models","arxiv_id":"2202.08906","date":"2022-02-17","proceeding":null,"authors":["Barret Zoph","Irwan Bello","Sameer Kumar","Nan Du","Yanping Huang","Jeff Dean","Noam Shazeer","William Fedus"],"abstract":"Scale has opened new frontiers in natural language processing -- but at a high cost. In response, Mixture-of-Experts (MoE) and Switch Transformers have been proposed as an energy efficient path to even larger and more capable language models. But advancing the state-of-the-art across a broad set of natural language tasks has been hindered by training instabilities and uncertain quality during fine-tuning. Our work focuses on these issues and acts as a design guide. We conclude by scaling a sparse model to 269B parameters, with a computational cost comparable to a 32B dense encoder-decoder Transformer (Stable and Transferable Mixture-of-Experts or ST-MoE-32B). For the first time, a sparse model achieves state-of-the-art performance in transfer learning, across a diverse set of tasks including reasoning (SuperGLUE, ARC Easy, ARC Challenge), summarization (XSum, CNN-DM), closed book question answering (WebQA, Natural Questions), and adversarially constructed tasks (Winogrande, ANLI R3).","url_abs":"https://arxiv.org/abs/2202.08906v2","url_pdf":"https://arxiv.org/pdf/2202.08906v2.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":"designing-effective-sparse-expert-models","repo_url":"https://github.com/tensorflow/mesh","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"tf","reach":null},{"paper_slug":"designing-effective-sparse-expert-models","repo_url":"https://github.com/xuefuzhao/openmoe","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"designing-effective-sparse-expert-models","repo_url":"https://github.com/yikangshen/megablocks","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"arc","task_name":"ARC"},{"task_slug":"common-sense-reasoning","task_name":"Common Sense Reasoning"},{"task_slug":"coreference-resolution","task_name":"Coreference Resolution"},{"task_slug":"decoder","task_name":"Decoder"},{"task_slug":"mixture-of-experts","task_name":"Mixture-of-Experts"},{"task_slug":"natural-language-inference","task_name":"Natural Language Inference"},{"task_slug":"natural-questions","task_name":"Natural Questions"},{"task_slug":"question-answering","task_name":"Question Answering"},{"task_slug":null,"task_name":"ST-MoE-L 4.1B (fine-tuned)"},{"task_slug":"transfer-learning","task_name":"Transfer Learning"},{"task_slug":"winogrande","task_name":"Winogrande"},{"task_slug":"word-sense-disambiguation","task_name":"Word Sense Disambiguation"}],"methods":[{"method_slug":"absolute-position-encodings","method_name":"Absolute Position Encodings"},{"method_slug":"adam","method_name":"Adam"},{"method_slug":"attention","method_name":"Attention"},{"method_slug":"bpe","method_name":"BPE"},{"method_slug":"dense-connections","method_name":"Dense Connections"},{"method_slug":"dropout","method_name":"Dropout"},{"method_slug":"label-smoothing","method_name":"Label Smoothing"},{"method_slug":"layer-normalization","method_name":"Layer Normalization"},{"method_slug":"linear-layer","method_name":"Linear Layer"},{"method_slug":"moe","method_name":"MoE"},{"method_slug":"multi-head-attention","method_name":"Multi-Head Attention"},{"method_slug":"position-wise-feed-forward-layer","method_name":"Position-Wise Feed-Forward Layer"},{"method_slug":"residual-connection","method_name":"Residual Connection"},{"method_slug":"softmax","method_name":"Softmax"},{"method_slug":"transformer","method_name":"Transformer"}],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/common-sense-reasoning-on-arc-challenge","task":"Common Sense Reasoning","dataset":"ARC (Challenge)","model":"ST-MoE-32B 269B (fine-tuned)","rank_in_archive_order":12,"of":54,"metrics":{"Accuracy":"86.5"},"uses_additional_data":false},{"leaderboard":"/sota/common-sense-reasoning-on-arc-challenge","task":"Common Sense Reasoning","dataset":"ARC (Challenge)","model":"ST-MoE-L 4.1B (fine-tuned)","rank_in_archive_order":28,"of":54,"metrics":{"Accuracy":"56.9"},"uses_additional_data":false},{"leaderboard":"/sota/common-sense-reasoning-on-arc-easy","task":"Common Sense Reasoning","dataset":"ARC (Easy)","model":"ST-MoE-32B 269B (fine-tuned)","rank_in_archive_order":1,"of":47,"metrics":{"Accuracy":"95.2"},"uses_additional_data":false},{"leaderboard":"/sota/common-sense-reasoning-on-arc-easy","task":"Common Sense Reasoning","dataset":"ARC (Easy)","model":"ST-MoE-L 4.1B (fine-tuned)","rank_in_archive_order":23,"of":47,"metrics":{"Accuracy":"75.4"},"uses_additional_data":false},{"leaderboard":"/sota/common-sense-reasoning-on-record","task":"Common Sense Reasoning","dataset":"ReCoRD","model":"ST-MoE-32B 269B (fine-tuned)","rank_in_archive_order":2,"of":45,"metrics":{"EM":"95.1"},"uses_additional_data":false},{"leaderboard":"/sota/common-sense-reasoning-on-record","task":"Common Sense Reasoning","dataset":"ReCoRD","model":"ST-MoE-L 4.1B (fine-tuned)","rank_in_archive_order":12,"of":45,"metrics":{"EM":"88.9"},"uses_additional_data":false},{"leaderboard":"/sota/common-sense-reasoning-on-winogrande","task":"Common Sense Reasoning","dataset":"WinoGrande","model":"ST-MoE-32B 269B (fine-tuned)","rank_in_archive_order":1,"of":77,"metrics":{"Accuracy":"96.1"},"uses_additional_data":false},{"leaderboard":"/sota/common-sense-reasoning-on-winogrande","task":"Common Sense Reasoning","dataset":"WinoGrande","model":"ST-MoE-L 4.1B (fine-tuned)","rank_in_archive_order":13,"of":77,"metrics":{"Accuracy":"81.7"},"uses_additional_data":false},{"leaderboard":"/sota/coreference-resolution-on-winograd-schema","task":"Coreference Resolution","dataset":"Winograd Schema Challenge","model":"ST-MoE-32B 269B (fine-tuned)","rank_in_archive_order":5,"of":82,"metrics":{"Accuracy":"96.6"},"uses_additional_data":false},{"leaderboard":"/sota/coreference-resolution-on-winograd-schema","task":"Coreference Resolution","dataset":"Winograd Schema Challenge","model":"ST-MoE-L 4.1B (fine-tuned)","rank_in_archive_order":8,"of":82,"metrics":{"Accuracy":"93.3"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-commitmentbank","task":"Natural Language Inference","dataset":"CommitmentBank","model":"ST-MoE-L 4.1B (fine-tuned)","rank_in_archive_order":3,"of":20,"metrics":{"Accuracy":"98.2"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-commitmentbank","task":"Natural Language Inference","dataset":"CommitmentBank","model":"ST-MoE-32B 269B (fine-tuned)","rank_in_archive_order":4,"of":20,"metrics":{"Accuracy":"98"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-rte","task":"Natural Language Inference","dataset":"RTE","model":"ST-MoE-32B 269B (fine-tuned)","rank_in_archive_order":4,"of":90,"metrics":{"Accuracy":"93.5%"},"uses_additional_data":false},{"leaderboard":"/sota/natural-language-inference-on-rte","task":"Natural Language Inference","dataset":"RTE","model":"ST-MoE-L 4.1B (fine-tuned)","rank_in_archive_order":10,"of":90,"metrics":{"Accuracy":"92.1%"},"uses_additional_data":false},{"leaderboard":"/sota/question-answering-on-boolq","task":"Question Answering","dataset":"BoolQ","model":"ST-MoE-32B 269B (fine-tuned)","rank_in_archive_order":2,"of":65,"metrics":{"Accuracy":"92.4"},"uses_additional_data":false},{"leaderboard":"/sota/question-answering-on-boolq","task":"Question Answering","dataset":"BoolQ","model":"ST-MoE-L 4.1B (fine-tuned)","rank_in_archive_order":11,"of":65,"metrics":{"Accuracy":"88.6"},"uses_additional_data":false},{"leaderboard":"/sota/question-answering-on-copa","task":"Question Answering","dataset":"COPA","model":"ST-MoE-32B 269B (fine-tuned)","rank_in_archive_order":3,"of":60,"metrics":{"Accuracy":"99.2"},"uses_additional_data":false},{"leaderboard":"/sota/question-answering-on-copa","task":"Question Answering","dataset":"COPA","model":"ST-MoE-L 4.1B (fine-tuned)","rank_in_archive_order":14,"of":60,"metrics":{"Accuracy":"91"},"uses_additional_data":false},{"leaderboard":"/sota/question-answering-on-multirc","task":"Question Answering","dataset":"MultiRC","model":"ST-MoE-32B 269B (fine-tuned)","rank_in_archive_order":2,"of":30,"metrics":{"F1":"89.6"},"uses_additional_data":false},{"leaderboard":"/sota/question-answering-on-multirc","task":"Question Answering","dataset":"MultiRC","model":"ST-MoE-L 4.1B (fine-tuned)","rank_in_archive_order":8,"of":30,"metrics":{"F1":"86"},"uses_additional_data":false},{"leaderboard":"/sota/word-sense-disambiguation-on-words-in-context","task":"Word Sense Disambiguation","dataset":"Words in Context","model":"ST-MoE-32B 269B (fine-tuned)","rank_in_archive_order":3,"of":37,"metrics":{"Accuracy":"77.7"},"uses_additional_data":false},{"leaderboard":"/sota/word-sense-disambiguation-on-words-in-context","task":"Word Sense Disambiguation","dataset":"Words in Context","model":"ST-MoE-L 4.1B (fine-tuned)","rank_in_archive_order":10,"of":37,"metrics":{"Accuracy":"74"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/2202.08906","atlas_url":"https://app.syntology.ai/?focus=2202.08906","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2202.08906"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-25T09:33:49+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/yikangshen/megablocks","reach":null},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/xuefuzhao/openmoe","reach":null},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/tensorflow/mesh","reach":null}],"summary":{"ran_draft_wrong":2,"ran_honours":3},"by_repo_kind":{"listed":{"samples":4,"ran":4,"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":5,"samples":[{"code_sha256_prefix":"90475ec2cfa11bd6","entry":"apply_llama_chat_template","repo":"xuefuzhao/openmoe","repo_kind":"listed","path":"script/inference_on_multi_devices.py","file_url":"https://github.com/xuefuzhao/openmoe/blob/HEAD/script/inference_on_multi_devices.py","link_basis":"first_harvest_node","language":"python","status":"ran_draft_wrong","verification_level":1,"contract_check":"OUTPUT_MISDECLARED","metamorphic_tier":"deterministic","behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"90475ec2cfa11bd6"}},{"code_sha256_prefix":"8eb924623b6d2a0f","entry":"calculate_num_tokens","repo":"xuefuzhao/openmoe","repo_kind":"listed","path":"eval/plot_bigbench.py","file_url":"https://github.com/xuefuzhao/openmoe/blob/HEAD/eval/plot_bigbench.py","link_basis":"first_harvest_node","language":"python","status":"ran_honours","verification_level":1,"contract_check":"HONOURS","metamorphic_tier":"well_formed","behaviour_fingerprint":true,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"8eb924623b6d2a0f"}},{"code_sha256_prefix":"3619322ac2eac1fa","entry":"convert_size","repo":"xuefuzhao/openmoe","repo_kind":"listed","path":"eval/plot_bigbench.py","file_url":"https://github.com/xuefuzhao/openmoe/blob/HEAD/eval/plot_bigbench.py","link_basis":"first_harvest_node","language":"python","status":"ran_honours","verification_level":1,"contract_check":"HONOURS","metamorphic_tier":"well_formed","behaviour_fingerprint":false,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"3619322ac2eac1fa"}},{"code_sha256_prefix":"7630a3eb1f9be798","entry":"promote_scalar","repo":null,"repo_kind":null,"path":null,"file_url":null,"link_basis":"identical_code_first_harvested_elsewhere","language":"python","status":"ran_honours","verification_level":1,"contract_check":"HONOURS","metamorphic_tier":"invariant","behaviour_fingerprint":true,"licence":null,"inline_ok":false,"mcp_get_code":{"code_sha256":"7630a3eb1f9be798"}},{"code_sha256_prefix":"6152639868e9cdb6","entry":"read_csv_and_convert_to_dataframe","repo":"xuefuzhao/openmoe","repo_kind":"listed","path":"eval/plot_bigbench.py","file_url":"https://github.com/xuefuzhao/openmoe/blob/HEAD/eval/plot_bigbench.py","link_basis":"first_harvest_node","language":"python","status":"ran_draft_wrong","verification_level":1,"contract_check":"OUTPUT_MISDECLARED","metamorphic_tier":"well_formed","behaviour_fingerprint":true,"licence":"NONE","inline_ok":false,"mcp_get_code":{"code_sha256":"6152639868e9cdb6"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}