add qwen3 to megatron conversion

[abdalgader-a]

IMPORTANT:
The NeMo submodule should be updated as in this PR: NVIDIA-NeMo/NeMo#14378

What does this PR do ?

This PR allow to export qwen3 model type from megatron format to HF

Test

Tested locally via run the megatron conversion script on Qwen3-1.7B.

Issues

List issues that this PR closes (syntax):

Usage

• You can potentially add a usage example below

# Add a code snippet demonstrating how to use this

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 run the unit tests and functional tests locally? Visit our Testing Guide for how to run tests
• Did you add or update any necessary documentation? Visit our Document Development Guide for how to write, build and test the docs.

Additional Information

• ...

[ashors1]

Thank you for the contribution! It looks like the NeMo submodule change touched a lot of unrelated files. Would it be possible to revert those unrelated changes and only keep the relevant ones?

[abdalgader-a]

Thanks @ashors1 for reviewing --- my apologies seems I mistakenly merged other changes. Now, I raised a new PR (#14378) in the NeMo submodule with only the relevant changes. Also, closed the wrong PR.

[ashors1]

Suggested change

	elif hf_config.model_type == "qwen3":
	elif hf_config.model_type in ("qwen3", "qwen3_moe"):

[ashors1]

Ah, actually I just tested this and it looks like this doesn't work with MoE models at the moment. Could we extend this to support MoE? If too much work, we could always add MoE support in a follow up PR

[ashors1]

@abdalgader-a let me know if you'd like to extend this PR to support Qwen3 MoE. I'm happy to help out if not

[abdalgader-a]

@ashors1 -- let me try to extend it. In case it takes long and much time needed we can raise another PR. I'll get back on this asap.

[ashors1]

@abdalgader-a I went ahead and started this process becuase we received some internal requests for the qwen3 moe exporter: https://github.com/NVIDIA-NeMo/RL/tree/ashors/qwen3-moe-export. I can raise the new PR and I'll be sure to add you as co-author.

[abdalgader-a]

now worries @ashors1! I can help out in the new PR too. I'll let you to choose either merge this PR or have them all in the new one.

[ashors1]

@abdalgader-a I went ahead and opened a new PR which adds MoE export support on top of your commits: #873. Please take a look. If this looks good to you, let's close this PR and focus on #873.

[abdalgader-a]

@ashors1 -- sorry for late reply. let's go ahead with #873. I'll close this PR.