Added eval scripts for VQA on the Clevr and Qlevr datasets

[jonathanlimsc]

Implemented eval scripts for VQA on the Clevr and Qlevr datasets, and tested to be working with model-path agi-collective/mistral-7b-oh-siglip-so400m-finetune-lora and model-base teknium/OpenHermes-2.5-Mistral-7B

Note that I assume the question file to be in the same format as found in the downloaded datasets.

Clevr

• Link: https://dl.fbaipublicfiles.com/clevr/CLEVR_v1.0.zip
• Question file CLEVR_v1.0/questions/CLEVR_test_questions.json , each question references an image found in CLEVR_v1.0/images/test.

Qlevr

• Link: https://drive.google.com/drive/folders/1s0n4CQXr1IDmBVUymYH51iC1MF54jyhN
• Question file questions_test.json, each question references an image with an id that corresponds to images in 2d_scene/test/full_images/2d_full_test_<image-id>.png or 3d_scene/test/images/3d_test_<image-id>.png.

[Alexis-BX]

Amazing thanks! Can you please remove unused imports please?