[{"scopus_import":"1","month":"08","abstract":[{"lang":"eng","text":"Elucidating reaction mechanisms requires efficient generation of transition states (TSs) and products. Existing diffusion and sequence-based models accelerate parts of this process over traditional string-based methods, but typically still require manual enumeration of either TSs or products, and stochastic diffusion dynamics can be inefficient and hard to control. We introduce MolGEN, a conditional flow-matching framework that uses deterministic optimal transport to map Gaussian priors to chemical distributions. For TS generation, MolGEN improves TS geometry and barrier-height prediction over diffusion models while enabling sub-second sampling. For reaction product generation, it achieves competitive top-k accuracy while preserving mass and electron balance. Using the same backbone for TS and product sampling, MolGEN enables template-free generative exploration of reaction networks without the repeated quantum-chemistry searches required by prior methods. For the γ-ketohydroperoxide decomposition network, it produces more valid TSs than string-based methods using only 12 quantum-chemistry evaluations instead of 1156, and identifies a lower-barrier pathway."}],"year":"2026","date_created":"2026-08-30T22:01:43Z","author":[{"last_name":"Tuo","id":"6e5644c0-c180-11ed-a2da-facc4c9f4f09","first_name":"Ping","full_name":"Tuo, Ping"},{"first_name":"Jiale","last_name":"Chen","id":"4d0a9064-1ff6-11ee-9fa6-ec046c604785","orcid":"0000-0001-5337-5875","full_name":"Chen, Jiale"},{"full_name":"Li, Ju","first_name":"Ju","last_name":"Li"}],"supplementarymaterial":"yes","external_id":{"pmid":["42469237"],"arxiv":["2507.10530"]},"intvolume":"        17","publication_status":"published","oa":1,"date_updated":"2026-09-09T07:28:15Z","OA_type":"gold","tmp":{"legal_code_url":"https://creativecommons.org/licenses/by/4.0/legalcode","name":"Creative Commons Attribution 4.0 International Public License (CC-BY 4.0)","short":"CC BY (4.0)","image":"/images/cc_by.png"},"quality_controlled":"1","language":[{"iso":"eng"}],"fulldoi":"https://doi.org/10.1038/s41467-026-75654-w","dataavailabilitystatement":"Structures generated in this study are provided in the Source Data file, and deposited in the GitHub repository https://github.com/tuoping/MolGEN and the figshare database under accession code https://doi.org/10.6084/m9.figshare.30576365. Source data are provided in this paper. The MolGEN codebase is available as an open-source repository for continuous development at https://github.com/tuoping/MolGEN. A release of the code used in this work has been archived on Zenodo47.","pmid":1,"file":[{"date_created":"2026-09-09T07:25:19Z","file_size":1526298,"file_id":"22864","file_name":"2026_NatureComm_Tuo.pdf","creator":"dernst","date_updated":"2026-09-09T07:25:19Z","relation":"main_file","checksum":"3cfb714304eec29b97dcaf7fb86dd6e3","content_type":"application/pdf","success":1,"access_level":"open_access"}],"user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","department":[{"_id":"DaAl"},{"_id":"GradSch"}],"article_type":"original","OA_place":"publisher","date_published":"2026-08-21T00:00:00Z","DOAJ_listed":"1","has_accepted_license":"1","acknowledgement":"P.T. thanks valued discussions with Dr. Peichen Zhong, Dr. Hao Tang, and Dr. Chengbin Zhao. P.T. thanks Dr. Dingshun Lv, Dr. Zechang Sun, and Dr. Chenxi Hu for identifying an important bug in an early version of the code package. The authors acknowledge the resources of the National Energy Research Scientific Computing Center (NERSC), a Department of Energy Office of Science User Facility using NERSC award DOEERCAP0031751 ‘GenAI@NERSC’. P.T. acknowledges funding from the BIDMaP Postdoctoral Fellowship.","doi":"10.1038/s41467-026-75654-w","day":"21","_id":"22771","publication":"Nature Communications","article_number":"8769","type":"journal_article","publication_identifier":{"eissn":["2041-1723"]},"file_date_updated":"2026-09-09T07:25:19Z","title":"Flow matching for reaction pathway generation","volume":17,"oa_version":"Published Version","arxiv":1,"ddc":["000"],"status":"public","article_processing_charge":"Yes","PlanS_conform":"1","citation":{"chicago":"Tuo, Ping, Jiale Chen, and Ju Li. “Flow Matching for Reaction Pathway Generation.” <i>Nature Communications</i>. Springer Nature, 2026. <a href=\"https://doi.org/10.1038/s41467-026-75654-w\">https://doi.org/10.1038/s41467-026-75654-w</a>.","ieee":"P. Tuo, J. Chen, and J. Li, “Flow matching for reaction pathway generation,” <i>Nature Communications</i>, vol. 17. Springer Nature, 2026.","apa":"Tuo, P., Chen, J., &#38; Li, J. (2026). Flow matching for reaction pathway generation. <i>Nature Communications</i>. Springer Nature. <a href=\"https://doi.org/10.1038/s41467-026-75654-w\">https://doi.org/10.1038/s41467-026-75654-w</a>","ama":"Tuo P, Chen J, Li J. Flow matching for reaction pathway generation. <i>Nature Communications</i>. 2026;17. doi:<a href=\"https://doi.org/10.1038/s41467-026-75654-w\">10.1038/s41467-026-75654-w</a>","ista":"Tuo P, Chen J, Li J. 2026. Flow matching for reaction pathway generation. Nature Communications. 17, 8769.","mla":"Tuo, Ping, et al. “Flow Matching for Reaction Pathway Generation.” <i>Nature Communications</i>, vol. 17, 8769, Springer Nature, 2026, doi:<a href=\"https://doi.org/10.1038/s41467-026-75654-w\">10.1038/s41467-026-75654-w</a>.","short":"P. Tuo, J. Chen, J. Li, Nature Communications 17 (2026)."},"researchdata_availability":"yes","publisher":"Springer Nature","das_tickbox":"1"},{"_id":"19877","day":"28","acknowledgement":"The authors would like to thank the Neural Magic team, in particular Michael Goin, Alexander Matveev, and Rob Shaw, for support with the vLLM integration. This research was supported in part by generous grants from NVIDIA and Google.","doi":"10.1145/3710848.3710871","has_accepted_license":"1","OA_place":"publisher","date_published":"2025-02-28T00:00:00Z","status":"public","article_processing_charge":"Yes (via OA deal)","arxiv":1,"ddc":["000"],"oa_version":"Published Version","isi":1,"file_date_updated":"2025-06-24T06:04:17Z","title":"MARLIN: Mixed-precision auto-regressive parallel inference on Large Language Models","publication_identifier":{"isbn":["9798400714436"]},"type":"conference","publication":"Proceedings of the 30th ACM SIGPLAN Annual Symposium on Principles and Practice of Parallel Programming","corr_author":"1","citation":{"chicago":"Frantar, Elias, Roberto L. Castro, Jiale Chen, Torsten Hoefler, and Dan-Adrian Alistarh. “MARLIN: Mixed-Precision Auto-Regressive Parallel Inference on Large Language Models.” In <i>Proceedings of the 30th ACM SIGPLAN Annual Symposium on Principles and Practice of Parallel Programming</i>, 239–51. Association for Computing Machinery, 2025. <a href=\"https://doi.org/10.1145/3710848.3710871\">https://doi.org/10.1145/3710848.3710871</a>.","ieee":"E. Frantar, R. L. Castro, J. Chen, T. Hoefler, and D.-A. Alistarh, “MARLIN: Mixed-precision auto-regressive parallel inference on Large Language Models,” in <i>Proceedings of the 30th ACM SIGPLAN Annual Symposium on Principles and Practice of Parallel Programming</i>, Las Vegas, NV, United States, 2025, pp. 239–251.","apa":"Frantar, E., Castro, R. L., Chen, J., Hoefler, T., &#38; Alistarh, D.-A. (2025). MARLIN: Mixed-precision auto-regressive parallel inference on Large Language Models. In <i>Proceedings of the 30th ACM SIGPLAN Annual Symposium on Principles and Practice of Parallel Programming</i> (pp. 239–251). Las Vegas, NV, United States: Association for Computing Machinery. <a href=\"https://doi.org/10.1145/3710848.3710871\">https://doi.org/10.1145/3710848.3710871</a>","short":"E. Frantar, R.L. Castro, J. Chen, T. Hoefler, D.-A. Alistarh, in:, Proceedings of the 30th ACM SIGPLAN Annual Symposium on Principles and Practice of Parallel Programming, Association for Computing Machinery, 2025, pp. 239–251.","mla":"Frantar, Elias, et al. “MARLIN: Mixed-Precision Auto-Regressive Parallel Inference on Large Language Models.” <i>Proceedings of the 30th ACM SIGPLAN Annual Symposium on Principles and Practice of Parallel Programming</i>, Association for Computing Machinery, 2025, pp. 239–51, doi:<a href=\"https://doi.org/10.1145/3710848.3710871\">10.1145/3710848.3710871</a>.","ama":"Frantar E, Castro RL, Chen J, Hoefler T, Alistarh D-A. MARLIN: Mixed-precision auto-regressive parallel inference on Large Language Models. In: <i>Proceedings of the 30th ACM SIGPLAN Annual Symposium on Principles and Practice of Parallel Programming</i>. Association for Computing Machinery; 2025:239-251. doi:<a href=\"https://doi.org/10.1145/3710848.3710871\">10.1145/3710848.3710871</a>","ista":"Frantar E, Castro RL, Chen J, Hoefler T, Alistarh D-A. 2025. MARLIN: Mixed-precision auto-regressive parallel inference on Large Language Models. Proceedings of the 30th ACM SIGPLAN Annual Symposium on Principles and Practice of Parallel Programming. PPoPP: Symposium on Principles and Practice of Parallel Programming, 239–251."},"page":"239-251","publisher":"Association for Computing Machinery","external_id":{"isi":["001437826500019"],"arxiv":["2408.11743"]},"author":[{"full_name":"Frantar, Elias","last_name":"Frantar","id":"09a8f98d-ec99-11ea-ae11-c063a7b7fe5f","first_name":"Elias"},{"last_name":"Castro","first_name":"Roberto L.","full_name":"Castro, Roberto L."},{"full_name":"Chen, Jiale","first_name":"Jiale","last_name":"Chen","id":"4d0a9064-1ff6-11ee-9fa6-ec046c604785","orcid":"0000-0001-5337-5875"},{"full_name":"Hoefler, Torsten","first_name":"Torsten","last_name":"Hoefler"},{"id":"4A899BFC-F248-11E8-B48F-1D18A9856A87","last_name":"Alistarh","orcid":"0000-0003-3650-940X","first_name":"Dan-Adrian","full_name":"Alistarh, Dan-Adrian"}],"date_created":"2025-06-23T13:51:58Z","year":"2025","conference":{"start_date":"2025-03-01","end_date":"2025-03-05","name":"PPoPP: Symposium on Principles and Practice of Parallel Programming","location":"Las Vegas, NV, United States"},"abstract":[{"lang":"eng","text":"As inference on Large Language Models (LLMs) emerges as an important workload in machine learning applications, model weight quantization has become a standard technique for efficient GPU deployment. Quantization not only reduces model size, but has also been shown to yield substantial speedups for single-user inference, due to reduced memory movement, with low accuracy impact. Yet, it remains a key open question whether speedups are achievable also in batched settings with multiple parallel clients, which are highly relevant for practical serving. It is unclear whether GPU kernels can be designed to remain practically memory-bound, while supporting the substantially increased compute requirements of batched workloads.\r\nIn this paper, we resolve this question positively by introducing a new design for Mixed-precision Auto-Regressive LINear kernels, called MARLIN. Concretely, given a model whose weights are compressed via quantization to, e.g., 4 bits per element, MARLIN shows that batchsizes up to 16-32 can be practically supported with close to maximum (4×) quantization speedup, and larger batchsizes up to 64-128 with gradually decreasing, but still significant, acceleration. MARLIN accomplishes this via a combination of techniques, such as asynchronous memory access, complex task scheduling and pipelining, and bespoke quantization support. Our experiments show that MARLIN's near-optimal performance on individual LLM layers across different scenarios can also lead to significant end-to-end LLM inference speedups (of up to 2.8×) when integrated with the popular vLLM open-source serving engine. Finally, we show that MARLIN is extensible to further compression techniques, like NVIDIA 2:4 sparsity, leading to additional speedups."}],"month":"02","scopus_import":"1","OA_type":"hybrid","tmp":{"legal_code_url":"https://creativecommons.org/licenses/by/4.0/legalcode","name":"Creative Commons Attribution 4.0 International Public License (CC-BY 4.0)","short":"CC BY (4.0)","image":"/images/cc_by.png"},"oa":1,"date_updated":"2025-09-30T13:41:57Z","publication_status":"published","fulldoi":"https://doi.org/10.1145/3710848.3710871","language":[{"iso":"eng"}],"quality_controlled":"1","related_material":{"record":[{"relation":"software","status":"public","id":"19884"}]},"department":[{"_id":"DaAl"}],"user_id":"317138e5-6ab7-11ef-aa6d-ffef3953e345","file":[{"checksum":"a0566ea3c168e8273501a5eb7d767cf8","content_type":"application/pdf","success":1,"access_level":"open_access","creator":"dernst","date_updated":"2025-06-24T06:04:17Z","relation":"main_file","file_id":"19883","file_name":"2025_PPoPP_Frantar.pdf","date_created":"2025-06-24T06:04:17Z","file_size":1330044}]},{"citation":{"mla":"Chen, Jiale, et al. “Scalable Mechanistic Neural Networks.” <i>13th International Conference on Learning Representations</i>, ICLR, 2025, pp. 63716–37.","ama":"Chen J, Yao D, Pervez AA, Alistarh D-A, Locatello F. Scalable mechanistic neural networks. In: <i>13th International Conference on Learning Representations</i>. ICLR; 2025:63716-63737.","short":"J. Chen, D. Yao, A.A. Pervez, D.-A. Alistarh, F. Locatello, in:, 13th International Conference on Learning Representations, ICLR, 2025, pp. 63716–63737.","ista":"Chen J, Yao D, Pervez AA, Alistarh D-A, Locatello F. 2025. Scalable mechanistic neural networks. 13th International Conference on Learning Representations. ICLR: International Conference on Learning Representations, 63716–63737.","ieee":"J. Chen, D. Yao, A. A. Pervez, D.-A. Alistarh, and F. Locatello, “Scalable mechanistic neural networks,” in <i>13th International Conference on Learning Representations</i>, Singapore, Singapore, 2025, pp. 63716–63737.","apa":"Chen, J., Yao, D., Pervez, A. A., Alistarh, D.-A., &#38; Locatello, F. (2025). Scalable mechanistic neural networks. In <i>13th International Conference on Learning Representations</i> (pp. 63716–63737). Singapore, Singapore: ICLR.","chicago":"Chen, Jiale, Dingling Yao, Adeel A Pervez, Dan-Adrian Alistarh, and Francesco Locatello. “Scalable Mechanistic Neural Networks.” In <i>13th International Conference on Learning Representations</i>, 63716–37. ICLR, 2025."},"corr_author":"1","page":"63716-63737","publisher":"ICLR","day":"01","_id":"20032","has_accepted_license":"1","OA_place":"publisher","date_published":"2025-04-01T00:00:00Z","arxiv":1,"ddc":["000"],"oa_version":"Published Version","article_processing_charge":"No","status":"public","type":"conference","publication_identifier":{"isbn":["9798331320850"]},"title":"Scalable mechanistic neural networks","file_date_updated":"2025-07-22T07:58:22Z","publication":"13th International Conference on Learning Representations","quality_controlled":"1","language":[{"iso":"eng"}],"related_material":{"link":[{"relation":"software","url":"https://github.com/IST-DASLab/ScalableMNN"}]},"user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","department":[{"_id":"DaAl"},{"_id":"FrLo"}],"file":[{"access_level":"open_access","success":1,"checksum":"64cfdb12ae3e4e8ba57b1403e1066776","content_type":"application/pdf","relation":"main_file","date_updated":"2025-07-22T07:58:22Z","creator":"dernst","file_name":"2025_ICLR_Chen.pdf","file_id":"20065","file_size":732745,"date_created":"2025-07-22T07:58:22Z"}],"author":[{"orcid":"0000-0001-5337-5875","id":"4d0a9064-1ff6-11ee-9fa6-ec046c604785","last_name":"Chen","first_name":"Jiale","full_name":"Chen, Jiale"},{"full_name":"Yao, Dingling","id":"d3e02e50-48a8-11ee-8f62-c108061797fa","last_name":"Yao","first_name":"Dingling"},{"first_name":"Adeel A","id":"fca6d90c-d47f-11ee-bc87-93ff51604981","last_name":"Pervez","full_name":"Pervez, Adeel A"},{"orcid":"0000-0003-3650-940X","last_name":"Alistarh","id":"4A899BFC-F248-11E8-B48F-1D18A9856A87","first_name":"Dan-Adrian","full_name":"Alistarh, Dan-Adrian"},{"first_name":"Francesco","last_name":"Locatello","id":"26cfd52f-2483-11ee-8040-88983bcc06d4","orcid":"0000-0002-4850-0683","full_name":"Locatello, Francesco"}],"external_id":{"arxiv":["2410.06074"]},"conference":{"start_date":"2025-04-24","end_date":"2025-04-28","name":"ICLR: International Conference on Learning Representations","location":"Singapore, Singapore"},"date_created":"2025-07-20T22:02:01Z","year":"2025","scopus_import":"1","month":"04","abstract":[{"text":"We propose Scalable Mechanistic Neural Network (S-MNN), an enhanced neural network framework designed for scientific machine learning applications involving long temporal sequences. By reformulating the original Mechanistic Neural Network (MNN) (Pervez et al., 2024), we reduce the computational time and space complexities from cubic and quadratic with respect to the sequence length, respectively, to linear. This significant improvement enables efficient modeling of long-term dynamics without sacrificing accuracy or interpretability. Extensive experiments demonstrate that S-MNN matches the original MNN in precision while substantially reducing computational resources. Consequently, S-MNN can drop-in replace the original MNN in applications, providing a practical and efficient tool for integrating mechanistic bottlenecks into neural network models of complex dynamical systems. Source code is available at https://github.com/IST-DASLab/ScalableMNN.","lang":"eng"}],"tmp":{"legal_code_url":"https://creativecommons.org/licenses/by/4.0/legalcode","name":"Creative Commons Attribution 4.0 International Public License (CC-BY 4.0)","short":"CC BY (4.0)","image":"/images/cc_by.png"},"OA_type":"diamond","date_updated":"2025-08-04T08:03:11Z","oa":1,"publication_status":"published"},{"isi":1,"publication_identifier":{"eissn":["1549-9626"],"issn":["1549-9618"]},"title":"Scalable multitemperature free energy sampling of classical Ising spin states","type":"journal_article","publication":"Journal of Chemical Theory and Computation","article_processing_charge":"No","status":"public","issue":"22","oa_version":"None","volume":21,"acknowledged_ssus":[{"_id":"ScienComp"}],"date_published":"2025-10-31T00:00:00Z","_id":"20704","day":"31","acknowledgement":"P.T. acknowledges funding from FFG MAGNIFICO and the BIDMaP Postdoctoral Fellowship. Z.Z. acknowledges funding from the European Union’s Horizon 2020 research and innovation program under the Marie Skłodowska-Curie grant agreement No. 101034413. The authors acknowledge the research computing facilities provided by the Institute of Science and Technology Austria (ISTA), and resources of the National Energy Research Scientific Computing Center (NERSC), a Department of Energy Office of Science User Facility using NERSC award DOEERCAP0031751 ’GenAI@NERSC’. P.T. acknowledges valued discussions with Dr. Daniel King, Dr. Lei Wang, and Dr. Fuzhi Dai.","doi":"10.1021/acs.jctc.5c01248","page":"11427-11435","publisher":"American Chemical Society","das_tickbox":"0","corr_author":"1","researchdata_availability":"no","citation":{"chicago":"Tuo, Ping, Zezhu Zeng, Jiale Chen, and Bingqing Cheng. “Scalable Multitemperature Free Energy Sampling of Classical Ising Spin States.” <i>Journal of Chemical Theory and Computation</i>. American Chemical Society, 2025. <a href=\"https://doi.org/10.1021/acs.jctc.5c01248\">https://doi.org/10.1021/acs.jctc.5c01248</a>.","ieee":"P. Tuo, Z. Zeng, J. Chen, and B. Cheng, “Scalable multitemperature free energy sampling of classical Ising spin states,” <i>Journal of Chemical Theory and Computation</i>, vol. 21, no. 22. American Chemical Society, pp. 11427–11435, 2025.","apa":"Tuo, P., Zeng, Z., Chen, J., &#38; Cheng, B. (2025). Scalable multitemperature free energy sampling of classical Ising spin states. <i>Journal of Chemical Theory and Computation</i>. American Chemical Society. <a href=\"https://doi.org/10.1021/acs.jctc.5c01248\">https://doi.org/10.1021/acs.jctc.5c01248</a>","ista":"Tuo P, Zeng Z, Chen J, Cheng B. 2025. Scalable multitemperature free energy sampling of classical Ising spin states. Journal of Chemical Theory and Computation. 21(22), 11427–11435.","ama":"Tuo P, Zeng Z, Chen J, Cheng B. Scalable multitemperature free energy sampling of classical Ising spin states. <i>Journal of Chemical Theory and Computation</i>. 2025;21(22):11427-11435. doi:<a href=\"https://doi.org/10.1021/acs.jctc.5c01248\">10.1021/acs.jctc.5c01248</a>","short":"P. Tuo, Z. Zeng, J. Chen, B. Cheng, Journal of Chemical Theory and Computation 21 (2025) 11427–11435.","mla":"Tuo, Ping, et al. “Scalable Multitemperature Free Energy Sampling of Classical Ising Spin States.” <i>Journal of Chemical Theory and Computation</i>, vol. 21, no. 22, American Chemical Society, 2025, pp. 11427–35, doi:<a href=\"https://doi.org/10.1021/acs.jctc.5c01248\">10.1021/acs.jctc.5c01248</a>."},"publication_status":"published","intvolume":"        21","OA_type":"closed access","date_updated":"2026-08-07T10:29:06Z","abstract":[{"lang":"eng","text":"Generative models have advanced significantly in sampling material systems with continuous variables, such as atomistic structures. However, their application to discrete variables, like atom types or spin states, remains underexplored. In this work, we introduce a discrete flow matching model, tailored for systems with discrete phase-space coordinates (e.g., the Ising model or a multicomponent system on a lattice). This approach enables a single model to sample free energy surfaces over a wide temperature range with minimal training overhead, and the model generation is scalable to larger lattice sizes than those in the training set. We demonstrate our approach on the 2D Ising model, showing efficient and reliable free energy sampling. These results highlight the potential of flow matching for low-cost, scalable free energy sampling in discrete systems and suggest promising extensions to alchemical degrees of freedom in crystalline materials. The codebase developed for this work is openly available at https://github.com/tuoping/alchemicalFES."}],"month":"10","scopus_import":"1","external_id":{"pmid":["41172130"],"isi":["001605927900001"]},"author":[{"full_name":"Tuo, Ping","id":"6e5644c0-c180-11ed-a2da-facc4c9f4f09","last_name":"Tuo","first_name":"Ping"},{"full_name":"Zeng, Zezhu","first_name":"Zezhu","orcid":"0000-0001-5126-4928","last_name":"Zeng","id":"54a2c730-803f-11ed-ab7e-95b29d2680e7"},{"orcid":"0000-0001-5337-5875","id":"4d0a9064-1ff6-11ee-9fa6-ec046c604785","last_name":"Chen","first_name":"Jiale","full_name":"Chen, Jiale"},{"full_name":"Cheng, Bingqing","first_name":"Bingqing","orcid":"0000-0002-3584-9632","last_name":"Cheng","id":"cbe3cda4-d82c-11eb-8dc7-8ff94289fcc9"}],"supplementarymaterial":"no","year":"2025","date_created":"2025-11-30T23:02:06Z","pmid":1,"related_material":{"link":[{"url":"https://github.com/tuoping/alchemicalFES","relation":"software"}]},"article_type":"original","department":[{"_id":"BiCh"},{"_id":"DaAl"}],"user_id":"317138e5-6ab7-11ef-aa6d-ffef3953e345","fulldoi":"https://doi.org/10.1021/acs.jctc.5c01248","ec_funded":1,"language":[{"iso":"eng"}],"quality_controlled":"1","project":[{"_id":"fc2ed2f7-9c52-11eb-aca3-c01059dda49c","call_identifier":"H2020","grant_number":"101034413","name":"IST-BRIDGE: International postdoctoral program"}]},{"OA_place":"repository","date_published":"2024-11-24T00:00:00Z","month":"11","abstract":[{"lang":"eng","text":"This is Marlin, a Mixed Auto-Regressive Linear kernel (and the name of one of the planet's fastest fish), an extremely optimized FP16xINT4 matmul kernel aimed at LLM inference that can deliver close to ideal (4x) speedups up to batchsizes of 16-32 tokens (in contrast to the 1-2 tokens of prior work with comparable speedup).\r\n\r\nAdditionally, it includes Sparse-Marlin, an extension of the MARLIN kernels adding support to 2:4 weight sparsity, achieving 5.3x speedups on NVIDIA GPUs (Ampere/Ada)."}],"has_accepted_license":"1","year":"2024","doi":"10.5281/ZENODO.14213091","date_created":"2025-06-24T06:09:18Z","day":"24","author":[{"first_name":"Elias","id":"09a8f98d-ec99-11ea-ae11-c063a7b7fe5f","last_name":"Frantar","full_name":"Frantar, Elias"},{"full_name":"Castro, Roberto","first_name":"Roberto","last_name":"Castro"},{"full_name":"Chen, Jiale","orcid":"0000-0001-5337-5875","last_name":"Chen","id":"4d0a9064-1ff6-11ee-9fa6-ec046c604785","first_name":"Jiale"},{"last_name":"Hoefler","first_name":"Torsten","full_name":"Hoefler, Torsten"},{"full_name":"Alistarh, Dan-Adrian","orcid":"0000-0003-3650-940X","id":"4A899BFC-F248-11E8-B48F-1D18A9856A87","last_name":"Alistarh","first_name":"Dan-Adrian"}],"_id":"19884","type":"research_data_reference","title":"MARLIN: Mixed-precision auto-regressive parallel inference on Large Language Models","main_file_link":[{"open_access":"1","url":"https://doi.org/10.5281/ZENODO.14213091"}],"oa":1,"date_updated":"2025-09-30T13:41:56Z","tmp":{"legal_code_url":"https://creativecommons.org/licenses/by/4.0/legalcode","name":"Creative Commons Attribution 4.0 International Public License (CC-BY 4.0)","short":"CC BY (4.0)","image":"/images/cc_by.png"},"ddc":["000"],"oa_version":"Published Version","article_processing_charge":"No","status":"public","fulldoi":"https://doi.org/10.5281/ZENODO.14213091","citation":{"ieee":"E. Frantar, R. Castro, J. Chen, T. Hoefler, and D.-A. Alistarh, “MARLIN: Mixed-precision auto-regressive parallel inference on Large Language Models.” Zenodo, 2024.","apa":"Frantar, E., Castro, R., Chen, J., Hoefler, T., &#38; Alistarh, D.-A. (2024). MARLIN: Mixed-precision auto-regressive parallel inference on Large Language Models. Zenodo. <a href=\"https://doi.org/10.5281/ZENODO.14213091\">https://doi.org/10.5281/ZENODO.14213091</a>","chicago":"Frantar, Elias, Roberto Castro, Jiale Chen, Torsten Hoefler, and Dan-Adrian Alistarh. “MARLIN: Mixed-Precision Auto-Regressive Parallel Inference on Large Language Models.” Zenodo, 2024. <a href=\"https://doi.org/10.5281/ZENODO.14213091\">https://doi.org/10.5281/ZENODO.14213091</a>.","mla":"Frantar, Elias, et al. <i>MARLIN: Mixed-Precision Auto-Regressive Parallel Inference on Large Language Models</i>. Zenodo, 2024, doi:<a href=\"https://doi.org/10.5281/ZENODO.14213091\">10.5281/ZENODO.14213091</a>.","ama":"Frantar E, Castro R, Chen J, Hoefler T, Alistarh D-A. MARLIN: Mixed-precision auto-regressive parallel inference on Large Language Models. 2024. doi:<a href=\"https://doi.org/10.5281/ZENODO.14213091\">10.5281/ZENODO.14213091</a>","ista":"Frantar E, Castro R, Chen J, Hoefler T, Alistarh D-A. 2024. MARLIN: Mixed-precision auto-regressive parallel inference on Large Language Models, Zenodo, <a href=\"https://doi.org/10.5281/ZENODO.14213091\">10.5281/ZENODO.14213091</a>.","short":"E. Frantar, R. Castro, J. Chen, T. Hoefler, D.-A. Alistarh, (2024)."},"corr_author":"1","publisher":"Zenodo","user_id":"2DF688A6-F248-11E8-B48F-1D18A9856A87","department":[{"_id":"DaAl"}],"related_material":{"record":[{"relation":"used_for_analysis_in","status":"public","id":"19877"}]}}]
