stanford-mast / infaas Goto Github PK

Hey thanks for the sharing codes, could you share your model variants?

I was trying to copy the public AMI (ami-036de08e2e59b4abc) mentioned in README in us-west-2, but I got

Failed to copy ami-036de08e2e59b4abc
You do not have permission to access the storage of this ami

See the screenshot below:

Hi,

May I know the public AMI mentioned in the README is gone or changed? I am not able to find it in all regions..

(ami-036de08e2e59b4abc) in us-west-2

Thank you!
Lam

Can I deploy or reproduce INFaaS on local server instead AWS? I am not available to AWS but I have the local GPU and CPU resource, can I deploy this architecture by docker or any other method?

Hi,

According to #11, the public AMI is gone, so I just skipped this step, and trying to build dependencies and INFaaS from scratch. I'm using a m5.2xlarge instance with Ubuntu 20.04 default AMI.

The thing is when I tried to build the GRPC, I got this compile error:

I've changed my enviroment to gcc5&g++5, and used exactly same cmake version/Redis. It looks like the newly installed version of GRPC has changed its source code/or there is a library confliction.

How can I resolve this, or can we get the public AMI with pre-build dependencies again?

Thanks!

Hi, I am trying to replicate some of your work.
But I have some questions regarding the variant generator.
Is it located in the source code, as I have not been able to find it?

In the case, it is not available I have the following questions regarding the generation of models in your work.
For GPU variants you used TensorRT.
For Neuron instances, you used the Neuron compiler to create different variants.
Did you generate model variants for CPU instances, and if so how?

It is stated that for the ResNet50 model INFaaS generates 50 variants.
How exactly did the generator end up with around 50 models?
If I am correct, you got 7 different variants based on changing batch size for each hardware platform.
For TensorRT you changed the datatype to FP16 if I am correct?
It would be very helpful to get a bit more of a detailed perspective on the generating of model variants, thank you very much.

Hi,

I am trying to setup INFaaS, and having problem with accessing the infaas-sample-public config database. Is the bucket still public?

Thanks,
Arpan

I am a first-year graduate student with limited knowledge of model serving. Currently, I can get a four-card GPU machine(I'm not sure I can get more😟). If I want to reproduce this paper, what should I do? If I can do it locally, I will share the results with you.🎇