Model code modifications

custom_got/modeling_GOT.py

@@ -558,7 +558,7 @@ class GOTQwenForCausalLM(Qwen2ForCausalLM):
 
         image_tensor_1 = image_processor_high(image)
 
-        input_ids = torch.as_tensor(inputs.input_ids).cuda()
+        input_ids = torch.as_tensor(inputs.input_ids)
 
         stop_str = conv.sep if conv.sep_style != SeparatorStyle.TWO else conv.sep2
         keywords = [stop_str]
@@ -566,7 +566,7 @@ class GOTQwenForCausalLM(Qwen2ForCausalLM):
         streamer = TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
 
         if stream_flag:
-            with torch.autocast("cuda", dtype=torch.bfloat16):
+            with torch.autocast("cpu", dtype=torch.bfloat16):
                 output_ids = self.generate(
                     input_ids,
                     images=[image_tensor_1.unsqueeze(0).half().cuda()],
@@ -578,7 +578,7 @@ class GOTQwenForCausalLM(Qwen2ForCausalLM):
                     stopping_criteria=[stopping_criteria]
                     )
         else:
-            with torch.autocast("cuda", dtype=torch.bfloat16):
+            with torch.autocast("cpu", dtype=torch.bfloat16):
                 output_ids = self.generate(
                     input_ids,
                     images=[image_tensor_1.unsqueeze(0).half().cuda()],

dataset_creation.py

@@ -3,11 +3,11 @@ import json
 import os
 
 
-dataset = pd.read_csv('ocr_task/data_80k/data.csv')
+dataset = pd.read_csv('data_80k/data.csv')
 labels = dataset['image_file']
 text = dataset['text']
 json_data = []
-images_path = 'drive/MyDrive/data_80k/output_images/'
+images_path = '/kaggle/input/hindi-english-images/data_80k/output_images/'
 for i in range(len(labels)):
     json_data.append(
         {