fix qwen2 vl failure in intel cpu and xpu

[sywangyi]

Hi, I see your guys have added qwen2 vl into tgi, and I try the model in intel cpu, and find it does not work, failure like

data: {"error":"Request failed during generation: Server error: output with shape [1] doesn't match the broadcast shape [873]","error_type":"generation"}

I debug it and find following code is missing in modeling and cause prefill output shape is not as expected.

@drbh @Narsil please help review it

[drbh]

hi @sywangyi thank you for sharing this fix! I've included these changes in another PR that includes a couple other tweaks/fixes to qwen2-vl. #2708. Once that PR is merged I'll close this one. Thank you again for the fix! 🙏

[sywangyi]

2708 is merged. so close the pr