fix

app.py

@@ -5,6 +5,7 @@ from transformers import (
     AutoProcessor,
     AutoModelForDocumentQuestionAnswering,
 )
+from transformers import pipeline
 import torch
 
 tokenizer_ru2en = AutoTokenizer.from_pretrained("Helsinki-NLP/opus-mt-ru-en")
@@ -21,6 +22,14 @@ git_model_base = AutoModelForDocumentQuestionAnswering.from_pretrained(
 
 device = "cuda" if torch.cuda.is_available() else "cpu"
 git_model_base.to(device)
+image_processor = git_processor_base.image_processor
+
+
+def preprocess_image(image):
+    """Преобразуем изображение для модели"""
+    image_rgb = image.convert("RGB")
+    processed = image_processor(image_rgb, return_tensors="pt").pixel_values.to(device)
+    return processed.pixel_values, processed.words, processed.boxes
 
 
 def translate_ru2en(text):
@@ -38,16 +47,22 @@ def translate_en2ru(text):
 
 
 def generate_answer_git(processor, model, image, question):
-    pixel_values = processor(images=image, return_tensors="pt").pixel_values.to(device)
-    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).to(device)
 
-    generated_ids = model.generate(
-        pixel_values=pixel_values, input_ids=input_ids, max_length=50
+    qa_pipeline = pipeline(
+        "document-question-answering",
+        model="andgrt/layoutlmv2-base-uncased_finetuned_docvqa",
     )
-    generated_answer = processor.batch_decode(generated_ids, skip_special_tokens=True)
-    return generated_answer[0]
+    return qa_pipeline(image, question)
+    # pixel_values, _, _ = preprocess_image(image)
+    # 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).to(device)
+
+    # generated_ids = model.generate(
+    #     pixel_values=pixel_values, input_ids=input_ids, max_length=50
+    # )
+    # generated_answer = processor.batch_decode(generated_ids, skip_special_tokens=True)
+    # return generated_answer[0]
 
 
 def generate_answer(image, question):