File size: 950 Bytes
815c641
f9ccb9d
d81d39e
 
 
db282a4
d81d39e
 
2dc4350
d81d39e
db282a4
 
 
d81d39e
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
import gradio as gr
import torch
from transformers import AutoProcessor, AutoModelForCausalLM
from huggingface_hub import hf_hub_download
from PIL import Image
from datasets import load_dataset

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

dataset = load_dataset("question","answers","image")

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

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

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))