Re: [Ecp-aurora-sdk-discuss] DL Framework modules available in the JLSE
Hi Becky, Thanks for this announcement. I am pleased ATS is finally supported. I have some questions that are useful to have answered publicly: 1. Is there any reason to use Preview-3? We had to wait so long for it that we get Preview-5 at the same time, which I assume is everything in Preview-3 + updates. 2. If we run on the ATS nodes on JLSE, what tools (similar to nvidia-smi) can be used to monitor gpu memory usage? 3. I tested CosmicTagger right away with Tensorflow, Preview-5. It has been hanging, before the first iteration, for more than 30 minutes. The page below indicates it runs successfully. How can we debug this? 4. What is the status of horovod in Preview-5? Thanks, and happy new year, Corey From: Ecp-aurora-sdk-discuss <[email protected]> on behalf of "Ramer, Rebecca via Ecp-aurora-sdk-discuss" <[email protected]> Reply-To: "Ramer, Rebecca" <[email protected]> Date: Thursday, January 14, 2021 at 10:53 AM To: "[email protected]" <[email protected]> Subject: [Ecp-aurora-sdk-discuss] DL Framework modules available in the JLSE Hi, We would like to announce availability of pytorch and tensorflow deep learning framework modules for use in the JLSE. The pytorch and tensorflow alpha 1 update ATS preview 3 modules provide python 3.6.10 python environments and use the oneapi beta09 version of the dpcpp compiler on iris, yarrow, and arcticus nodes. More details, including how to setup and use the python environments can be viewed on the wiki, see: https://wiki.jlse.anl.gov/display/inteldga/Alpha1+update+-+ATS+Preview+3+Pyt... and https://wiki.jlse.anl.gov/display/inteldga/Alpha1+update+-+ATS+Preview+3+Ten... The alpha 1 update ATS preview 5 tensorflow and pytorch modules provide python 3.6.10 python environments and use the oneapi gold nda version of the dpcpp compiler on iris, yarrow, and arcticus nodes. Details, including how to setup and use the python environments can be viewed on the wiki, see: https://wiki.jlse.anl.gov/display/inteldga/Alpha1+update+-+ATS+Preview+5+Pyt... and https://wiki.jlse.anl.gov/display/inteldga/Alpha1+update+-+ATS+Preview+5+Ten... Any feedback is welcome. Rebecca Ramer [email protected]
participants (1)
-
Adams, Corey James