search

UofT-EcoSystem / hfta

Boost hardware utilization for ML training workloads via Inter-model Horizontal Fusion
https://UofT-EcoSystem.github.io/hfta/
MIT License
32 stars 11 forks source link

[Docker] Upload Docker Images for Reproducing MLSys'21 Artifact to DockerHub #24

Closed wangshangsam closed 3 years ago

wangshangsam commented 3 years ago

Just in case if people don't wanna rebuild those images again.

wangshangsam commented 3 years ago

native1.6-cu10.2:

docker pull wangshangsam/hfta:mlsys21_native1.6-cu10.2
wangshangsam commented 3 years ago

nvidia20.06

docker pull wangshangsam/hfta:mlsys21_nvidia20.06
wangshangsam commented 3 years ago

@nixli @ypm1999 I'm wondering if you could help to check if those two images work, then update docs/mlsys21/a100_throughputs.md and docs/mlsys21/v100_rtx6000_throughputs.md accordinly (just say something like "if you don't want to build the image manually, you can pull our pre-built image by xxx")?

ypm1999 commented 3 years ago

I will check it

xinli-git commented 3 years ago

I am checking rtx6000, if you have not checked a100 yet, I can do that as well :)

ypm1999 commented 3 years ago

I am checking A100, it works well up now

xinli-git commented 3 years ago

same for RTX6000, I'll update the README

wangshangsam commented 3 years ago

@ypm1999 I don't see docker pull wangshangsam/hfta:mlsys21_native1.6-cu10.2 being mentioned in #27 .

@nixli I'm wondering if you could update https://github.com/UofT-EcoSystem/hfta/blob/main/docs/mlsys21/a100_throughputs.md#build-docker-image as well?

xinli-git commented 3 years ago

added in #34