Image-Text-to-Text
sentence-transformers
Safetensors
Transformers
qwen2_vl
Qwen2-VL
conversational
cheesyFishes commited on
Commit
d6e8ec9
·
verified ·
1 Parent(s): 4474bfe

fix device handling

Browse files
Files changed (1) hide show
  1. custom_st.py +1 -5
custom_st.py CHANGED
@@ -27,7 +27,6 @@ class Transformer(nn.Module):
27
  tokenizer_args: Optional[Dict[str, Any]] = None,
28
  config_args: Optional[Dict[str, Any]] = None,
29
  cache_dir: Optional[str] = None,
30
- device: str = 'cpu',
31
  backend: Literal['torch', 'onnx', 'openvino'] = 'torch',
32
  **kwargs,
33
  ) -> None:
@@ -38,7 +37,6 @@ class Transformer(nn.Module):
38
  f'Backend \'{backend}\' is not supported, please use \'torch\' instead'
39
  )
40
 
41
- self.device = device
42
  self.dimension = dimension
43
  self.max_pixels = max_pixels
44
  self.min_pixels = min_pixels
@@ -160,15 +158,13 @@ class Transformer(nn.Module):
160
  def tokenize(self, texts: List[Union[str, Image.Image]], padding: str = 'longest') -> Dict[str, torch.Tensor]:
161
  processed_texts, processed_images = self._process_input(texts)
162
 
163
- inputs = self.processor(
164
  text=processed_texts,
165
  images=processed_images,
166
  videos=None,
167
  padding=padding,
168
  return_tensors='pt'
169
  )
170
-
171
- return {k: v.to(self.device) for k, v in inputs.items()}
172
 
173
  def save(self, output_path: str, safe_serialization: bool = True) -> None:
174
  """Save the model, tokenizer and processor to the given path."""
 
27
  tokenizer_args: Optional[Dict[str, Any]] = None,
28
  config_args: Optional[Dict[str, Any]] = None,
29
  cache_dir: Optional[str] = None,
 
30
  backend: Literal['torch', 'onnx', 'openvino'] = 'torch',
31
  **kwargs,
32
  ) -> None:
 
37
  f'Backend \'{backend}\' is not supported, please use \'torch\' instead'
38
  )
39
 
 
40
  self.dimension = dimension
41
  self.max_pixels = max_pixels
42
  self.min_pixels = min_pixels
 
158
  def tokenize(self, texts: List[Union[str, Image.Image]], padding: str = 'longest') -> Dict[str, torch.Tensor]:
159
  processed_texts, processed_images = self._process_input(texts)
160
 
161
+ return self.processor(
162
  text=processed_texts,
163
  images=processed_images,
164
  videos=None,
165
  padding=padding,
166
  return_tensors='pt'
167
  )
 
 
168
 
169
  def save(self, output_path: str, safe_serialization: bool = True) -> None:
170
  """Save the model, tokenizer and processor to the given path."""