{"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/mixlora-enhancing-large-language-models-fine","title":"MixLoRA: Enhancing Large Language Models Fine-Tuning with LoRA-based Mixture of Experts","arxiv_id":"2404.15159","date":"2024-04-22","proceeding":null,"authors":["Dengchun Li","Yingzi Ma","Naizheng Wang","Zhengmao Ye","Zhiyuan Cheng","Yinghao Tang","Yan Zhang","Lei Duan","Jie Zuo","Cal Yang","Mingjie Tang"],"abstract":"Fine-tuning Large Language Models (LLMs) is a common practice to adapt pre-trained models for specific applications. While methods like LoRA have effectively addressed GPU memory constraints during fine-tuning, their performance often falls short, especially in multi-task scenarios. In contrast, Mixture-of-Expert (MoE) models, such as Mixtral 8x7B, demonstrate remarkable performance in multi-task learning scenarios while maintaining a reduced parameter count. However, the resource requirements of these MoEs remain challenging, particularly for consumer-grade GPUs with less than 24GB memory. To tackle these challenges, we propose MixLoRA, an approach to construct a resource-efficient sparse MoE model based on LoRA. MixLoRA inserts multiple LoRA-based experts within the feed-forward network block of a frozen pre-trained dense model and employs a commonly used top-k router. Unlike other LoRA-based MoE methods, MixLoRA enhances model performance by utilizing independent attention-layer LoRA adapters. Additionally, an auxiliary load balance loss is employed to address the imbalance problem of the router. Our evaluations show that MixLoRA improves about 9% accuracy compared to state-of-the-art PEFT methods in multi-task learning scenarios. We also propose a new high-throughput framework to alleviate the computation and memory bottlenecks during the training and inference of MOE models. This framework reduces GPU memory consumption by 40% and token computation latency by 30% during both training and inference.","url_abs":"https://arxiv.org/abs/2404.15159v3","url_pdf":"https://arxiv.org/pdf/2404.15159v3.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":"mixlora-enhancing-large-language-models-fine","repo_url":"https://github.com/TUDB-Labs/MixLoRA","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"Apache-2.0"}},{"paper_slug":"mixlora-enhancing-large-language-models-fine","repo_url":"https://github.com/mikecovlee/mLoRA","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"Apache-2.0"}}],"tasks":[{"task_slug":"common-sense-reasoning","task_name":"Common Sense Reasoning"},{"task_slug":null,"task_name":"GPU"},{"task_slug":"mixture-of-experts","task_name":"Mixture-of-Experts"},{"task_slug":"multi-task-learning","task_name":"Multi-Task Learning"},{"task_slug":"quantization","task_name":"Quantization"},{"task_slug":"question-answering","task_name":"Question Answering"},{"task_slug":"sentence-completion","task_name":"Sentence Completion"},{"task_slug":"text-classification","task_name":"Text Classification"}],"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":"mixlora","method_name":"MixLoRA"},{"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":[{"slug":"mixlora","name":"MixLoRA","full_name":"MixLoRA"}],"results":[{"leaderboard":"/sota/common-sense-reasoning-on-arc-challenge","task":"Common Sense Reasoning","dataset":"ARC (Challenge)","model":"LLaMA-3 8B + MixLoRA","rank_in_archive_order":17,"of":54,"metrics":{"Accuracy":"79.9"},"uses_additional_data":false},{"leaderboard":"/sota/common-sense-reasoning-on-arc-challenge","task":"Common Sense Reasoning","dataset":"ARC (Challenge)","model":"LLaMA-2 13B + MixLoRA","rank_in_archive_order":18,"of":54,"metrics":{"Accuracy":"69.9"},"uses_additional_data":false},{"leaderboard":"/sota/common-sense-reasoning-on-arc-challenge","task":"Common Sense Reasoning","dataset":"ARC (Challenge)","model":"LLaMA-2 7B + MixLoRA","rank_in_archive_order":26,"of":54,"metrics":{"Accuracy":"58.1"},"uses_additional_data":false},{"leaderboard":"/sota/common-sense-reasoning-on-arc-easy","task":"Common Sense Reasoning","dataset":"ARC (Easy)","model":"LLaMA-3 8B + MixLoRA","rank_in_archive_order":5,"of":47,"metrics":{"Accuracy":"86.5"},"uses_additional_data":false},{"leaderboard":"/sota/common-sense-reasoning-on-arc-easy","task":"Common Sense Reasoning","dataset":"ARC (Easy)","model":"LLaMA-2 13B + MixLoRA","rank_in_archive_order":10,"of":47,"metrics":{"Accuracy":"83.5"},"uses_additional_data":false},{"leaderboard":"/sota/common-sense-reasoning-on-arc-easy","task":"Common Sense Reasoning","dataset":"ARC (Easy)","model":"LLaMA-2 7B + MixLoRA","rank_in_archive_order":20,"of":47,"metrics":{"Accuracy":"77.7"},"uses_additional_data":false},{"leaderboard":"/sota/common-sense-reasoning-on-winogrande","task":"Common Sense Reasoning","dataset":"WinoGrande","model":"LLaMA-2 13B + MixLoRA","rank_in_archive_order":9,"of":77,"metrics":{"Accuracy":"86.3"},"uses_additional_data":false},{"leaderboard":"/sota/common-sense-reasoning-on-winogrande","task":"Common Sense Reasoning","dataset":"WinoGrande","model":"LLaMA-3 8B + MixLoRA","rank_in_archive_order":12,"of":77,"metrics":{"Accuracy":"82.1"},"uses_additional_data":false},{"leaderboard":"/sota/common-sense-reasoning-on-winogrande","task":"Common Sense Reasoning","dataset":"WinoGrande","model":"LLaMA-2 7B + MixLoRA","rank_in_archive_order":24,"of":77,"metrics":{"Accuracy":"76.8"},"uses_additional_data":false},{"leaderboard":"/sota/question-answering-on-boolq","task":"Question Answering","dataset":"BoolQ","model":"LLaMA-2 13B + MixLoRA","rank_in_archive_order":31,"of":65,"metrics":{"Accuracy":"77.1"},"uses_additional_data":false},{"leaderboard":"/sota/question-answering-on-boolq","task":"Question Answering","dataset":"BoolQ","model":"LLaMA-3 8B + MixLoRA","rank_in_archive_order":36,"of":65,"metrics":{"Accuracy":"75"},"uses_additional_data":false},{"leaderboard":"/sota/question-answering-on-boolq","task":"Question Answering","dataset":"BoolQ","model":"LLaMA-2 7B + MixLoRA","rank_in_archive_order":40,"of":65,"metrics":{"Accuracy":"72.7"},"uses_additional_data":false},{"leaderboard":"/sota/question-answering-on-openbookqa","task":"Question Answering","dataset":"OpenBookQA","model":"LLaMA-3 8B + MixLoRA","rank_in_archive_order":16,"of":45,"metrics":{"Accuracy":"84.8"},"uses_additional_data":false},{"leaderboard":"/sota/question-answering-on-openbookqa","task":"Question Answering","dataset":"OpenBookQA","model":"LLaMA-2 13B + MixLoRA","rank_in_archive_order":19,"of":45,"metrics":{"Accuracy":"83"},"uses_additional_data":false},{"leaderboard":"/sota/question-answering-on-openbookqa","task":"Question Answering","dataset":"OpenBookQA","model":"LLaMA-2 7B + MixLoRA","rank_in_archive_order":24,"of":45,"metrics":{"Accuracy":"81.6"},"uses_additional_data":false},{"leaderboard":"/sota/question-answering-on-piqa","task":"Question Answering","dataset":"PIQA","model":"LLaMA-3 8B + MixLoRA","rank_in_archive_order":4,"of":67,"metrics":{"Accuracy":"87.6"},"uses_additional_data":false},{"leaderboard":"/sota/question-answering-on-piqa","task":"Question Answering","dataset":"PIQA","model":"LLaMA-2 13B + MixLoRA","rank_in_archive_order":7,"of":67,"metrics":{"Accuracy":"86.8"},"uses_additional_data":false},{"leaderboard":"/sota/question-answering-on-piqa","task":"Question Answering","dataset":"PIQA","model":"LLaMA-2 7B + MixLoRA","rank_in_archive_order":15,"of":67,"metrics":{"Accuracy":"83.2"},"uses_additional_data":false},{"leaderboard":"/sota/question-answering-on-social-iqa","task":"Question Answering","dataset":"SIQA","model":"LLaMA-2 13B + MixLoRA","rank_in_archive_order":2,"of":24,"metrics":{"Accuracy":"82.5"},"uses_additional_data":false},{"leaderboard":"/sota/question-answering-on-social-iqa","task":"Question Answering","dataset":"SIQA","model":"LLaMA-3 8B + MixLoRA","rank_in_archive_order":10,"of":24,"metrics":{"Accuracy":"78.8"},"uses_additional_data":false},{"leaderboard":"/sota/question-answering-on-social-iqa","task":"Question Answering","dataset":"SIQA","model":"LLaMA-2 7B + MixLoRA","rank_in_archive_order":11,"of":24,"metrics":{"Accuracy":"78"},"uses_additional_data":false},{"leaderboard":"/sota/sentence-completion-on-hellaswag","task":"Sentence Completion","dataset":"HellaSwag","model":"LLaMA-2 13B + MixLoRA","rank_in_archive_order":7,"of":89,"metrics":{"Accuracy":"94.7"},"uses_additional_data":false},{"leaderboard":"/sota/sentence-completion-on-hellaswag","task":"Sentence Completion","dataset":"HellaSwag","model":"LLaMA-3 8B + MixLoRA","rank_in_archive_order":9,"of":89,"metrics":{"Accuracy":"93.3"},"uses_additional_data":false},{"leaderboard":"/sota/sentence-completion-on-hellaswag","task":"Sentence Completion","dataset":"HellaSwag","model":"LLaMA-2 7B + MixLoRA","rank_in_archive_order":10,"of":89,"metrics":{"Accuracy":"93.1"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/2404.15159","atlas_url":"https://app.syntology.ai/?focus=2404.15159","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2404.15159"}},"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/mikecovlee/mLoRA","reach":{"status":"ok","spdx":"Apache-2.0"}},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/TUDB-Labs/MixLoRA","reach":{"status":"ok","spdx":"Apache-2.0"}}],"summary":{"ran":6,"unverified":5},"by_repo_kind":{"official":{"samples":11,"ran":6,"repositories":2}},"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":"0a82ce127c214e4b","entry":"apply_rotary_pos_emb","repo":"mikecovlee/mLoRA","repo_kind":"official","path":"mlora/models/modeling_chatglm.py","file_url":"https://github.com/mikecovlee/mLoRA/blob/HEAD/mlora/models/modeling_chatglm.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"0a82ce127c214e4b"}},{"code_sha256_prefix":"5b93021aaf3e8b4c","entry":"dequantize_module_weight","repo":"TUDB-Labs/MixLoRA","repo_kind":"official","path":"mixlora/lora_linear.py","file_url":"https://github.com/TUDB-Labs/MixLoRA/blob/HEAD/mixlora/lora_linear.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"5b93021aaf3e8b4c"}},{"code_sha256_prefix":"45a78242fc74dcd9","entry":"is_package_available","repo":"TUDB-Labs/MixLoRA","repo_kind":"official","path":"mixlora/utils.py","file_url":"https://github.com/TUDB-Labs/MixLoRA/blob/HEAD/mixlora/utils.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"45a78242fc74dcd9"}},{"code_sha256_prefix":"22f4910fcb84030d","entry":"logits_process","repo":"mikecovlee/mLoRA","repo_kind":"official","path":"mlora/generator.py","file_url":"https://github.com/mikecovlee/mLoRA/blob/HEAD/mlora/generator.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"22f4910fcb84030d"}},{"code_sha256_prefix":"9281708eb4cf9161","entry":"split_tensor_along_last_dim","repo":"mikecovlee/mLoRA","repo_kind":"official","path":"mlora/models/modeling_chatglm.py","file_url":"https://github.com/mikecovlee/mLoRA/blob/HEAD/mlora/models/modeling_chatglm.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"9281708eb4cf9161"}},{"code_sha256_prefix":"7f59a97bd91fdef5","entry":"swiglu","repo":"mikecovlee/mLoRA","repo_kind":"official","path":"mlora/models/modeling_chatglm.py","file_url":"https://github.com/mikecovlee/mLoRA/blob/HEAD/mlora/models/modeling_chatglm.py","link_basis":"first_harvest_node","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":true,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"7f59a97bd91fdef5"}},{"code_sha256_prefix":"602bfa722a261a61","entry":"apply_partial_rotary_emb","repo":"mikecovlee/mLoRA","repo_kind":"official","path":"mlora/models/modeling_phi.py","file_url":"https://github.com/mikecovlee/mLoRA/blob/HEAD/mlora/models/modeling_phi.py","link_basis":"first_harvest_node","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":"602bfa722a261a61"}},{"code_sha256_prefix":"3ce1a041a1abc119","entry":"dequantize_bnb_weight","repo":"TUDB-Labs/MixLoRA","repo_kind":"official","path":"mixlora/lora_linear.py","file_url":"https://github.com/TUDB-Labs/MixLoRA/blob/HEAD/mixlora/lora_linear.py","link_basis":"first_harvest_node","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":"3ce1a041a1abc119"}},{"code_sha256_prefix":"ae2f069de9355ede","entry":"load_adapter_weights","repo":"TUDB-Labs/MixLoRA","repo_kind":"official","path":"mixlora/model.py","file_url":"https://github.com/TUDB-Labs/MixLoRA/blob/HEAD/mixlora/model.py","link_basis":"first_harvest_node","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":"ae2f069de9355ede"}},{"code_sha256_prefix":"949439173c840737","entry":"load_dataset","repo":"mikecovlee/mLoRA","repo_kind":"official","path":"mlora/dispatcher.py","file_url":"https://github.com/mikecovlee/mLoRA/blob/HEAD/mlora/dispatcher.py","link_basis":"first_harvest_node","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":"949439173c840737"}},{"code_sha256_prefix":"4cd248daacbead2f","entry":"query_yes_no","repo":"mikecovlee/mLoRA","repo_kind":"official","path":"mlora.py","file_url":"https://github.com/mikecovlee/mLoRA/blob/HEAD/mlora.py","link_basis":"first_harvest_node","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":"4cd248daacbead2f"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}