[Feature]: phi-3.5 is a strong model for its size, including vision support. Has multi-image support, but vllm does not support

[pseudotensor]

🚀 The feature, motivation and pitch

phi-3.5 is a strong model for its size, including strong multi-image vision support. But vllm does not support the multi-image case.

vllm/vllm/model_executor/models/phi3v.py

Lines 421 to 425 in 03b7bfb

	elif len(re.findall(r"(<\|image_\d+\|>)+", prompt)) > 1:
	logger.warning("Multiple image input is not supported yet, "
	"so any extra image tokens will be treated "
	"as plain text.")

Alternatives

Only other models

Additional context

No response

[DarkLight1337]

@Isotr0py are you interested in implementing this?

[Isotr0py]

Of course, I'm just working on implementing this feature.
I will create a PR once it's nearly to be finished.