CODEWITHSUNDEEP
tensorflowdeploymentgoogle-cloud-platform
Aniel Rossi
Aniel Rossi

Reputation: 3

Error --accelerator unrecognized argument when launching gcloud beta ai-platform versions create API

I would like to deploy my Tensorflow model on a Google AI Platform machine. I already have a working version on a legacy machine mls1-c4-m2. Now I would like to make the prediction faster and use a n1-standard-4 machine with a K80 Nvidia GPU.

I am following this guide: https://cloud.google.com/ai-platform/prediction/docs/machine-types-online-prediction

I have instantiated a model in us-central1, as described in the guide.

This is the command:

gcloud beta ai-platform versions create $VERSION \
 --model $MODEL \
 --origin $ORIGIN \
 --runtime-version 1.15 \
 --python-version 3.7 \
 --framework tensorflow \
 --machine-type n1-standard-4 \
 --accelerator count=1,type=nvidia-tesla-k80 \
 --config $CONFIG

This is the configuration file:

runtimeVersion: '1.15'
manualScaling:
  nodes: 1

This is the error that I get:

ERROR: (gcloud.beta.ai-platform.versions.create) unrecognized arguments:
--accelerator
count=1,type=nvidia-tesla-k80

Has anyone experienced this?

Upvotes: 0

Views: 410

Answers (2)

Swarna Bharathi
Swarna Bharathi

Reputation: 1

In my case, it was the extra tab before --accelerator flag. Make sure formatting is correct.

Thanks to @Nicholas Elkaim's comment above, as I did not suspect/notice it.

Upvotes: -1

William.P
William.P

Reputation: 124

It seems that the --accelerator flag is not recognized. Maybe you need to upgrade gcloud components

Upvotes: 1

Related Questions