Andrey VelichkevichandGitHub e051fe09da KEP-2170: [SDK] Initial implementation of the Kubeflow Training V2 Python SDK (#2324) ...

* Init commit

Signed-off-by: Andrey Velichkevich <[email protected]>

* Implement SDK APIs

Signed-off-by: Andrey Velichkevich <[email protected]>

* Update __init__ file

Signed-off-by: Andrey Velichkevich <[email protected]>

* Add model and dataset configs

Signed-off-by: Andrey Velichkevich <[email protected]>

* Remove print

Signed-off-by: Andrey Velichkevich <[email protected]>

* Add status to TrainJob

Signed-off-by: Andrey Velichkevich <[email protected]>

* Fix sed for Linux

Signed-off-by: Andrey Velichkevich <[email protected]>

* Fix PHASE_POST_TRAINING const

Signed-off-by: Andrey Velichkevich <[email protected]>

* Add device labels for the runtimes

Signed-off-by: Andrey Velichkevich <[email protected]>

* Add Components to the TrainJob type

Signed-off-by: Andrey Velichkevich <[email protected]>

* Get container devices util

Signed-off-by: Andrey Velichkevich <[email protected]>

* Use SDK assets in initializer

Signed-off-by: Andrey Velichkevich <[email protected]>

* Add the runtime_ref to the list_jobs() API

Signed-off-by: Andrey Velichkevich <[email protected]>

* Check if runtime label presents

Signed-off-by: Andrey Velichkevich <[email protected]>

* Add Trainer arg as part of train API

Signed-off-by: Andrey Velichkevich <[email protected]>

* Rename lora to peft config

Signed-off-by: Andrey Velichkevich <[email protected]>

* Using global prop to remove SDK tests

Signed-off-by: Andrey Velichkevich <[email protected]>

* Fix type for lora_dropout

Signed-off-by: Andrey Velichkevich <[email protected]>

* Import the JobSet models
Use kube-openapi generator to remove unnecessary models from apimachinery

Signed-off-by: Andrey Velichkevich <[email protected]>

* Parse runtime as object

Signed-off-by: Andrey Velichkevich <[email protected]>

* Remove old gen_openapi

Signed-off-by: Andrey Velichkevich <[email protected]>

* Rename GPU_DEVICE_LABEL to NVIDIA_GPU_DEVICE_LABEL

Signed-off-by: Andrey Velichkevich <[email protected]>

* Run codegen

Signed-off-by: Andrey Velichkevich <[email protected]>

* Install git on initializer images

Signed-off-by: Andrey Velichkevich <[email protected]>

* Rename device to accelerator for Runtime class

Signed-off-by: Andrey Velichkevich <[email protected]>

---------

Signed-off-by: Andrey Velichkevich <[email protected]>

2024-12-13 16:59:06 +00:00

35 KiB

Raw Blame History