from transformers import AutoProcessor, AutoModelForCausalLM
from huggingface_hub import hf_hub_download
from PIL import Image

processor = AutoProcessor.from_pretrained("microsoft/git-base-vqav2")
model = AutoModelForCausalLM.from_pretrained("microsoft/git-base-vqav2")

file_path = hf_hub_download(repo_id="Multimodal-Fatima/OK-VQA_train", filename="data", repo_type="dataset")
image = Image.open(file_path).convert("RGB")

pixel_values = processor(images=image, return_tensors="pt").pixel_values

question = "How many people are there?"

input_ids = processor(text=question, add_special_tokens=False).input_ids
input_ids = [processor.tokenizer.cls_token_id] + input_ids
input_ids = torch.tensor(input_ids).unsqueeze(0)

generated_ids = model.generate(pixel_values=pixel_values, input_ids=input_ids, max_length=50)
print(processor.batch_decode(generated_ids, skip_special_tokens=True))