Support Qwen2.5-7B-Instruct model

[tfogal]

🚀 Model / language coverage

Support the Qwen/Qwen2.5-7B-Instruct model.

Pitch

This is an ask from internal NVIDIA colleagues.

Minimal Repro

TBD.

cc @tfogal

[tfogal]

The NeMo team simplified this to a non-VL variant of the model.

[IvanYashchuk]

#1406 adds a non-VL variant of the Qwen 2 model to Thunder's CI. Currently backward doesn't work because of NVIDIA/Fuser#871 (comment).

We still need to continue with performance analysis and microbenchmarks to improve parts of Qwen 2 (and other two important HF models Phi 3, NeMo Mistral) that do work with nvFuser fusing backward operations. @riccardofelluga, could you please lead this effort? It would also be useful to identify what layer causes nvFuser to fail for the backward computation.