Update app.py

app.py

@@ -31,40 +31,26 @@ git_model_base.to(device)
 # vilt_model.to(device)
 
 def generate_answer_git(processor, model, image, question):
-    # # prepare image
-    # pixel_values = processor(images=image, return_tensors="pt").pixel_values
+    # prepare image
+    pixel_values = processor(images=image, return_tensors="pt").pixel_values
 
-    # # prepare question
-    # 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)
+    # prepare question
+    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, return_dict_in_generate=True, output_scores=True)
-    # print('scores:')
-    # print(generated_ids.scores)
-    # idx = generated_ids.scores[0].argmax(-1).item()
-    # idx1 = generated_ids.scores[1].argmax(-1).item()
-    # print(idx, idx1)
-    # print(model.config.id2label)
-    # ans = model.config.id2label[idx]
-    # ans1 = model.config.id2label[idx1]
-    # print(ans, ans1)
-    # print('sequences:')
-    # print(generated_ids.sequences)
-    # print(generated_ids)
-    # generated_answer = processor.batch_decode(generated_ids.sequences, skip_special_tokens=True)
-    # print(generated_answer)
-
-    encoding = processor(images=image, text=question, return_tensors="pt")
-
-    with torch.no_grad():
-        outputs = model(**encoding)
-    print(outputs.logits)
-    predicted_class_idx = outputs.logits[0].argmax(-1).item()
-    # return model.config.id2label[predicted_class_idx]
-    print(predicted_class_idx)
+    generated_ids = model.generate(pixel_values=pixel_values, input_ids=input_ids, max_length=50, return_dict_in_generate=True, output_scores=True)
+    print('scores:')
+    print(generated_ids.scores)
+    idx = generated_ids.scores[0].argmax(-1).item()
+    idx1 = generated_ids.scores[1].argmax(-1).item()
+    print(idx, idx1)
     print(model.config.id2label)
-    print(model.config.id2label[predicted_class_idx])
+    print('sequences:')
+    print(generated_ids.sequences)
+    print(generated_ids)
+    generated_answer = processor.batch_decode(generated_ids.sequences, skip_special_tokens=True)
+    print(generated_answer)
     
    
     return 'haha'