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

Flyte

Source: <https://www.union.ai/docs/flyte/tutorials/model-training/mnist-classifier/#distributed-data-parallel-training>
> Flyte also supports distributed training for PyTorch models, but this is not native. It is achieved using {ref}`kf-pytorch-op`, for example.
what does this mean that its not native? is there a tutorial for this?

The non native refers to it uses an external plugin that you have to install 

<@U04H9F0NW3E> we should say it’s not default 

<@U08MUMQRV9C> you have to install the KF training operator 

Also <@U04H9F0NW3E> - I wouldn’t say not native. I would say - needs additional dependencies. Also it seems the ref is not working 

Are there any public examples showing how to use Kubeflow Trainer with Flyte?

<@U08MUMQRV9C> sorry but something is wrong with docs rendering - we are looking

Here is raw code <https://github.com/flyteorg/flytesnacks/blob/c9b2c44dc5d0ce42e482be1ffb16c8b19e72c23c/examples/kfpytorch_plugin/kfpytorch_plugin/pytorch_mnist.py#L204|https://github.com/flyteorg/flytesnacks/blob/c9b2c44dc5d0ce42e482be1ffb16c8b19e72c23c/examples/kfpytorch_plugin/kfpytorch_plugin/pytorch_mnist.py#L204>

Let me look at what’s happening with the docs

So <@U08MUMQRV9C> basically idea is everything is @task and if you want special plugin - you can develop anything you use task_config=…

For example Ray, Spark, dask, distributed training, sagemaker, slurm etc

<@U08MUMQRV9C> actually the doc is there - just the link is weirdly missing from the DDP page - <https://www.union.ai/docs/flyte/integrations/native-backend-plugins/kfpytorch_plugin/pytorch_mnist/>

which version of KF Training Operator does Flyte support? V1 or V2 or both?

<https://www.kubeflow.org/docs/components/trainer/legacy-v1/>

Reading the code here: <https://github.com/flyteorg/flytekit/blob/master/plugins/flytekit-kf-pytorch/flytekitplugins/kfpytorch/task.py>

It seems to point to the deprecated training-operator repository: <https://github.com/kubeflow/pytorch-operator>

Does it still work with the latest release of KF Training Operator V1?

Ah, finally found the correct docs page: <https://www.union.ai/docs/flyte/deployment/flyte-plugins/kubernetes-plugins/>