Explicitly set dtype.

requirements.txt

@@ -2,8 +2,8 @@
 phonemizer==3.3.0
 transformers==4.45.2
 gradio-client==1.4.0
-torch==2.0.1
-torchaudio==2.0.2
+torch==2.1.0
+torchaudio==2.1.0
 torchvision==0.16.0
 spaces==0.30.3
 huggingface-hub==0.25.1

src/model/modules/imagecraftprocessor.py

@@ -63,7 +63,7 @@ class ImageCraftProcessor:
         # Convert the list of numpy arrays to a single numpy array with shape [Batch_Size, Channel, Height, Width]
         pixel_values = np.stack(pixel_values, axis=0)
         # Convert the numpy array to a PyTorch tensor
-        pixel_values = torch.tensor(pixel_values)
+        pixel_values = torch.tensor(pixel_values, dtype=torch.float16)
 
         input_strings = [
             add_image_tokens_to_prompt(