Flyte enables production-grade orchestration for machine learning workflows and data processing created to accelerate local workflows to production.

Flyte

Hello Team,

I want to specify a GPU pool for my map task. I went through this <https://docs.flyte.org/projects/flytekit/en/latest/generated/flytekit.Resources.html#flytekit.Resources.gpu|documentation> and it is not very clear. Could you please help with this ?

hi <@U03HWQ0RPFV> have you seen <https://docs.flyte.org/projects/cookbook/en/latest/auto/deployment/configure_use_gpus.html#sphx-glr-auto-deployment-configure-use-gpus-py>?

Hi <@UPBBNMXD1> - We use this spark config to specify in which pool to launch a spark task in

`"<http://spark.kubernetes.node.selector.l5.lyft.com/pool|spark.kubernetes.node.selector.l5.lyft.com/pool>": eks-pdx-pool-gpu,`


Now, <@U03HWQ0RPFV> is trying to convert the spark task to a map task which requires GPU, we are wondering how to do that.

Should we add it to task’s Resource?

you should definitely add GPU to the task resource decorator but please go through the above documentation (specifically with regards to the configmap changes) in order for the map tasks to be scheduled on machines with gpus

Sure <@UPBBNMXD1> = Thank you, we will look into that!

&gt; you should definitely add GPU to the task resource decorator
Can you show an example for this?

&gt; please go through the above documentation
Where should we add that config? in propeller/admin ?

Bear with me for all noob questions! :slightly_smiling_face:

no problem, we love questions :smile: (and finding ways to improve our documentation) you'll need to update the propeller configmap

for adding gpu to the map task instances, you can use the with_overrides syntax: <https://docs.flyte.org/projects/flytekit/en/latest/generated/flytekit.map_task.html#flytekit-map-task>

```k8s:
        co-pilot:
          name: "flyte-copilot-"
          image: ${ecr_repository}:flytecopilot-${flytecopilot_version}
          start-timeout: "30s"
        scheduler-name: flyte-scheduler
        inject-finalizer: true
        default-annotations:
          - <http://flyte.lyft.com/deployment|flyte.lyft.com/deployment>: flyte-l5
        resource-tolerations:
          <http://nvidia.com/gpu|nvidia.com/gpu>:
            key: <http://lyft.com/gpu|lyft.com/gpu>
            value: dedicated
            operator: Equal
            effect: NoSchedule```

Looks like we already have this added to our propeller config Map :slightly_smiling_face:

We launched Map tasks on gpu successfully!

How can we edit/update flyte documentation?

<@U0283D92TUK>, here’s a detailed <https://docs.flyte.org/projects/cookbook/en/latest/contribute.html#contribute-to-documentation|guide> to update docs. In a nutshell,

• Fork the `flytesnacks` repo
• Update the relevant Python file
• `pip install -r docs-requirements.txt` in the `cookbook` folder
• Run `make html` in the `docs` folder
• Open HTML pages present in the `docs/_build` directory in the browser to verify if everything’s rendered correctly