no code implementations • 26 Feb 2024 • Maximilian Böther, Abraham Sebastian, Pranjal Awasthi, Ana Klimovic, Srikumar Ramalingam
In this paper, we relax the requirement of having a central machine for the target subset by proposing a novel distributed bounding algorithm with provable approximation guarantees.
no code implementations • 11 Dec 2023 • Maximilian Böther, Viktor Gsteiger, Ties Robroek, Ana Klimovic
Machine learning training data is often dynamic in real-world use cases, i. e., data is added or removed and may experience distribution shifts over time.
1 code implementation • 8 Dec 2023 • Xiaozhe Yao, Ana Klimovic
Fine-tuning large language models (LLMs) for downstream tasks can greatly improve model quality, however serving many different fine-tuned LLMs concurrently for users in multi-tenant environments is challenging.
no code implementations • 26 Oct 2022 • Andrew Audibert, Yang Chen, Dan Graur, Ana Klimovic, Jiri Simsa, Chandramohan A. Thekkath
To avoid data stalls, the host CPU and RAM required for input data processing per accelerator core used for ML computations varies across jobs.
1 code implementation • 4 Apr 2022 • Cedric Renggli, Xiaozhe Yao, Luka Kolar, Luka Rimanic, Ana Klimovic, Ce Zhang
Transfer learning can be seen as a data- and compute-efficient alternative to training models from scratch.
2 code implementations • 7 Nov 2021 • Michael Kuchnik, Ana Klimovic, Jiri Simsa, Virginia Smith, George Amvrosiadis
Our analysis of over two million ML jobs in Google datacenters reveals that a significant fraction of model training jobs could benefit from faster input data pipelines.
1 code implementation • 17 May 2021 • Jiawei Jiang, Shaoduo Gan, Yue Liu, Fanlin Wang, Gustavo Alonso, Ana Klimovic, Ankit Singla, Wentao Wu, Ce Zhang
The appeal of serverless (FaaS) has triggered a growing interest on how to use it in data-intensive applications such as ETL, query processing, or machine learning (ML).
no code implementations • 28 Jan 2021 • Derek G. Murray, Jiri Simsa, Ana Klimovic, Ihor Indyk
Finally, we characterize machine learning input pipelines for millions of jobs that ran in Google's fleet, showing that input data processing is highly diverse and consumes a significant fraction of job resources.