Reconsider pre-training and post-training phases for the Training Runtimes

[andreyvelich]

What you would like to be added?

Currently, we categorize our runtimes as either pre-training or post-training.
The original intent was to place LLM Fine-Tuning blueprints under post-training runtimes while reserving pre-training for runtimes without an initializer or a pre-configured LLM trainer.

However, this distinction may be confusing for users since pre-training runtimes can also be used for Fine-Tuning—the only difference being that users need to create a training function with a Fine-Tuning script manually.

To improve clarity, we should explore a better way to differentiate these two types of Training Runtimes.

cc @kubeflow/wg-training-leads @Electronic-Waste @astefanutti @shravan-achar @akshaychitneni @saileshd1402 @deepanker13

Why is this needed?

We need to make it clear how to select Training Runtime.

Love this feature?

Give it a 👍 We prioritize the features with most 👍