{"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/omni-smola-boosting-generalist-multimodal","title":"Omni-SMoLA: Boosting Generalist Multimodal Models with Soft Mixture of Low-rank Experts","arxiv_id":"2312.00968","date":"2023-12-01","proceeding":"CVPR 2024 1","authors":["Jialin Wu","Xia Hu","Yaqing Wang","Bo Pang","Radu Soricut"],"abstract":"Large multi-modal models (LMMs) exhibit remarkable performance across numerous tasks. However, generalist LMMs often suffer from performance degradation when tuned over a large collection of tasks. Recent research suggests that Mixture of Experts (MoE) architectures are useful for instruction tuning, but for LMMs of parameter size around O(50-100B), the prohibitive cost of replicating and storing the expert models severely limits the number of experts we can use. We propose Omni-SMoLA, an architecture that uses the Soft MoE approach to (softly) mix many multimodal low rank experts, and avoids introducing a significant number of new parameters compared to conventional MoE models. The core intuition here is that the large model provides a foundational backbone, while different lightweight experts residually learn specialized knowledge, either per-modality or multimodally. Extensive experiments demonstrate that the SMoLA approach helps improve the generalist performance across a broad range of generative vision-and-language tasks, achieving new SoTA generalist performance that often matches or outperforms single specialized LMM baselines, as well as new SoTA specialist performance.","url_abs":"https://arxiv.org/abs/2312.00968v2","url_pdf":"https://arxiv.org/pdf/2312.00968v2.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":[],"tasks":[{"task_slug":"chart-question-answering","task_name":"Chart Question Answering"},{"task_slug":"document-ai","task_name":"Document AI"},{"task_slug":"image-captioning","task_name":"Image Captioning"},{"task_slug":"mixture-of-experts","task_name":"Mixture-of-Experts"},{"task_slug":"object-counting","task_name":"Object Counting"},{"task_slug":"visual-question-answering","task_name":"Visual Question Answering (VQA)"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/chart-question-answering-on-chartqa","task":"Chart Question Answering","dataset":"ChartQA","model":"SMoLA-PaLI-X Specialist Model","rank_in_archive_order":7,"of":27,"metrics":{"1:1 Accuracy":"74.6"},"uses_additional_data":true},{"leaderboard":"/sota/chart-question-answering-on-chartqa","task":"Chart Question Answering","dataset":"ChartQA","model":"SMoLA-PaLI-X Generalist Model","rank_in_archive_order":8,"of":27,"metrics":{"1:1 Accuracy":"73.8"},"uses_additional_data":true},{"leaderboard":"/sota/object-counting-on-tallyqa-complex","task":"Object Counting","dataset":"TallyQA-Complex","model":"SMoLA-PaLI-X Specialist","rank_in_archive_order":1,"of":6,"metrics":{"Accuracy":"77.1"},"uses_additional_data":true},{"leaderboard":"/sota/object-counting-on-tallyqa-complex","task":"Object Counting","dataset":"TallyQA-Complex","model":"SMoLA-PaLI-X Generalist (0 shot)","rank_in_archive_order":3,"of":6,"metrics":{"Accuracy":"70.7"},"uses_additional_data":true},{"leaderboard":"/sota/object-counting-on-tallyqa-simple","task":"Object Counting","dataset":"TallyQA-Simple","model":"SMoLA-PaLI-X Specialist","rank_in_archive_order":1,"of":6,"metrics":{"Accuracy":"86.3"},"uses_additional_data":true},{"leaderboard":"/sota/object-counting-on-tallyqa-simple","task":"Object Counting","dataset":"TallyQA-Simple","model":"SMoLA-PaLI-X Generalist (0 shot)","rank_in_archive_order":3,"of":6,"metrics":{"Accuracy":"83.3"},"uses_additional_data":true},{"leaderboard":"/sota/visual-question-answering-on-a-okvqa","task":"Visual Question Answering (VQA)","dataset":"A-OKVQA","model":"SMoLA-PaLI-X Specialist Model","rank_in_archive_order":1,"of":15,"metrics":{"DA VQA Score":"70.55","MC Accuracy":"83.75"},"uses_additional_data":true},{"leaderboard":"/sota/visual-question-answering-vqa-on-ai2d","task":"Visual Question Answering (VQA)","dataset":"AI2D","model":"SMoLA-PaLI-X Specialist Model","rank_in_archive_order":1,"of":4,"metrics":{"EM":"82.5"},"uses_additional_data":true},{"leaderboard":"/sota/visual-question-answering-vqa-on-ai2d","task":"Visual Question Answering (VQA)","dataset":"AI2D","model":"SMoLA-PaLI-X Generalist Model","rank_in_archive_order":2,"of":4,"metrics":{"EM":"81.4"},"uses_additional_data":true},{"leaderboard":"/sota/visual-question-answering-on-docvqa-test","task":"Visual Question Answering (VQA)","dataset":"DocVQA test","model":"SMoLA-PaLI-X Specialist","rank_in_archive_order":3,"of":33,"metrics":{"ANLS":"0.908"},"uses_additional_data":true},{"leaderboard":"/sota/visual-question-answering-on-docvqa-test","task":"Visual Question Answering (VQA)","dataset":"DocVQA test","model":"SMoLA-PaLI-X Generalist","rank_in_archive_order":4,"of":33,"metrics":{"ANLS":"0.906"},"uses_additional_data":true},{"leaderboard":"/sota/visual-question-answering-vqa-on","task":"Visual Question Answering (VQA)","dataset":"InfographicVQA","model":"SMoLA-PaLI-X Specialist","rank_in_archive_order":2,"of":21,"metrics":{"ANLS":"66.2"},"uses_additional_data":true},{"leaderboard":"/sota/visual-question-answering-vqa-on","task":"Visual Question Answering (VQA)","dataset":"InfographicVQA","model":"SMoLA-PaLI-X Generalist","rank_in_archive_order":4,"of":21,"metrics":{"ANLS":"65.6"},"uses_additional_data":true}],"syntology":{"syntology_url":"https://syntology.ai/paper/2312.00968","atlas_url":"https://app.syntology.ai/?focus=2312.00968","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}