feat: input validation & model capability

[akoumpa]

What does this PR do ?

Example:

>> model = NeMoAutoModelForCausalLM.from_pretrained("nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16")
>> print(model.supports_generate)
>> True

Introduces a model capabilities mixin, which allows users to query a model instance for the following attributes:
model.

• supports_tp
• supports_pp
• supports_cp
• supports_ep
• supports_tp_plan # HF compat
• supports_pp_plan # HF compat
• supports_sequence_packing
• supports_gradient_checkpointing
• supports_cp_with_sequence_packing
• supports_generate
• is_custom_model

The above information is used together with current device_mesh configuration to validate whether the input parallelism configuration is valid or not.

Additionally, at the train loop level, we can check for things like context parallelism + sequence packing support.

Changelog

• Add specific line by line info of high level changes in this PR.

Before your PR is "Ready for review"

Pre checks:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests?
• Did you add or update any necessary documentation?

If you haven't finished some of the above items you can still open "Draft" PR.

Additional Information

• Related to # (issue)

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[akoumpa]

/ok to test 6a6fe26

[akoumpa]

/ok to test 125ebdc

[akoumpa]

/ok to test aa9aec1

[akoumpa]

/ok to test 2fb9709

[akoumpa]

/ok to test 4a32efd

[akoumpa]

/ok to test 7da29da

[akoumpa]

/ok to test ac100ab

[akoumpa]

/ok to test ab0746d

[akoumpa]

/claude review

[akoumpa]

/ok to test 654b9ad