Support for the Qwen3 models in engine-flow

[michaelfeil]

Similar PRs:

• [Model] Adding Qwen3 and Qwen3MoE sgl-project/sglang#4693
• Adding Qwen3 and Qwen3MoE huggingface/transformers#36878

We are most interested in the regular QWen (non moe) model. Happy to provide additonal details.

Edit: Now that the model is released, some more additional details:

• it would be great to have the qwen3 model in the quantize_and_export workflow with fp8 precision
• support for the modeling code of engine in qwen3 and qwen3_moe

[michaelfeil]

I have such implementation working internally. If someone from Nvidia will manage the review, @basetenlabs would be happy to contribute this for the engine flow.

[plume]

@michaelfeil I'm really looking forward to this. Let's do our best!