ML Reproducibility Challenge 2023 on Chameleon – Sign Up

ML Reproducibility Challenge 2023 on Chameleon

Welcome to our application page. Fill out the form below to indicate your interest in using Chameleon resources to support participants in the ML Reproducibility Challenge 2023! After you fill out the form, a member of our team will reach out to you to explain how to receive an allocation on Chameleon for your team.

Please review your application thoroughly before submission. Completed forms should be submitted by filling out the form below. Only those who are eligible to receive PI status on Chameleon should apply (see our guidelines for PI eligibility here). For any queries regarding the application process or resource allocation, please contact mtrichardson@uchicago.edu.

Chameleon Cloud charges user projects in Service Units, which equate to one hour of wall clock time on a single bare metal server. We charge SUs based on each resource you reserve for a project, including computer servers, storage, GPU and FPGA nodes, and network resources (floating IPs and VLANs). You can read more about Service Unit (SU) charges and the different multiplies associated with certain resources (e.g., GPUs) by reading our FAQs. You can also see an example of an SU cost estimation for an ML project below.

Thank you for your interest in the ML Reproducibility Challenge and for contributing to the advancement of machine learning research through educational collaboration and resource sharing.

SU Cost Estimation Example for an ML Project on Chameleon

  • Skylake: 10 servers, 10 FIPs, 0 networks reserved (VLANs), 10 hours, SU Calculation (10 + 10) * 10, 200 SUs needed.
  • GPUv100: 1 server, 1 FIP, 1 network reserved (VLAN), 10 hours, SU Calculation (2 + 1 + 1) * 10, 40 SUs needed.
  • storage_nvme: 1 server, 1 FIP, 1 network reserved (VLAN), 10 hours, SU Calculation (2 + 1 + 1) * 10, 40 SUs needed.
  • FPGA: 1 server, 0 FIPs, 0 networks reserved (VLANs), 10 hours, SU Calculation (2 + 0 + 0) * 10, 20 SUs needed.
Please provide links to each paper.
As a simple example, if a student's experiment requires a bare metal server with a GPU and a public IP address for SSH access, this setup would cost 3 SUs for each hour that it is reserved and would cost in total 24 SUs for an 8-hour experiment run. Please see the information in the form description above for more advice on SU usage.
Do you have preferred compute resources for your team?
Have you previously used Chameleon resources for other research or projects?
I certify that the information provided is accurate and that the resources, if allocated, will be used solely for the purpose of the ML Reproducibility Challenge as outlined in the project proposals. I understand that the allocation of resources is subject to review and approval based on the feasibility and relevance of the submitted projects.

This form is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.