Image Text To Text Support

[IlyasMoutawwakil]

this is a bit more complicated than other tasks because the inputs are not the same across model types (and backends), for example the inputs are the usual input_ids and pixel_values with model types such as blip, blip2, git, ... but are more complex for model types such as idefics, idefics2, qwen2_vl, ... because they handle multiple images in the same prompt with interleaved text and image tokens.

• idefics
• idefics2
• qwen2_vl
• generic image-text-to-text (blip, blip2, git)

[IlyasMoutawwakil]

tests run on cpu and are passing for multiple architectures, if any custom input is needed, don't hesitate to open an issue/pr.
failing tests are unrelated.