{"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/fleetx","title":"FleetX","arxiv_id":null,"date":"2021-12-06","proceeding":"arXiv 2021 12","authors":["EW"],"abstract":"Distributed training has become a pervasive and effective approach for training a large neural network\r\n(NN) model with processing massive data. However, it is very challenging to satisfy requirements\r\nfrom various NN models, diverse computing resources, and their dynamic changes during a training\r\njob. In this study, we design our distributed training framework in a systematic end-to-end view to\r\nprovide the built-in adaptive ability for different scenarios, especially for industrial applications and\r\nproduction environments, by fully considering resource allocation, model partition, task placement,\r\nand distributed execution. Based on the unified distributed graph and the unified cluster object,\r\nour adaptive framework is equipped with a global cost model and a global planner, which can\r\nenable arbitrary parallelism, resource-aware placement, multi-mode execution, fault-tolerant, and\r\nelastic distributed training. The experiments demonstrate that our framework can satisfy various\r\nrequirements from the diversity of applications and the heterogeneity of resources with highly\r\ncompetitive performance. The ERNIE language model with 260 billion parameters is efficiently\r\ntrained on thousands of AI processors with 91.7% weak scalability. The throughput of the model\r\nfrom the recommender system by employing the heterogeneous pipeline asynchronous execution can\r\nbe increased up to 2.1 times and 3.3 times that of the GPU-only and CPU-only training respectively.\r\nMoreover, the fault-tolerant and elastic distributed training have been successfully applied to the\r\nonline industrial applications, which give a reduction of 34.49% in the number of failed long-term\r\ntraining jobs and an increase of 33.91% for the global scheduling efficiency in the production\r\nenvironment.","url_abs":"https://github.com/PaddlePaddle/FleetX","url_pdf":"https://github.com/PaddlePaddle/FleetX","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":"fleetx","repo_url":"https://github.com/PaddlePaddle/FleetX","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"paddle","reach":{"status":"ok","spdx":"Apache-2.0"}}],"tasks":[{"task_slug":null,"task_name":"CPU"},{"task_slug":"distributed-computing","task_name":"Distributed Computing"},{"task_slug":null,"task_name":"GPU"},{"task_slug":"language-modelling","task_name":"Language Modelling"},{"task_slug":"recommendation-systems","task_name":"Recommendation Systems"},{"task_slug":"scheduling","task_name":"Scheduling"}],"methods":[{"method_slug":"ernie","method_name":"ERNIE"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}