Fix: use torch.from_numpy() to create tensors for np.ndarrays (#33201)

use torch.from_numpy for np.ndarrays

src/transformers/agents/agent_types.py

@@ -105,7 +105,7 @@ def __init__(self, value):
         elif isinstance(value, torch.Tensor):
             self._tensor = value
         elif isinstance(value, np.ndarray):
-            self._tensor = torch.tensor(value)
+            self._tensor = torch.from_numpy(value)
         else:
             raise TypeError(f"Unsupported type for {self.__class__.__name__}: {type(value)}")
 
@@ -192,7 +192,10 @@ def __init__(self, value, samplerate=16_000):
             self._tensor = value
         elif isinstance(value, tuple):
             self.samplerate = value[0]
-            self._tensor = torch.tensor(value[1])
+            if isinstance(value[1], np.ndarray):
+                self._tensor = torch.from_numpy(value[1])
+            else:
+                self._tensor = torch.tensor(value[1])
         else:
             raise ValueError(f"Unsupported audio type: {type(value)}")
 

src/transformers/agents/document_question_answering.py

@@ -60,7 +60,7 @@ def encode(self, document: "Image", question: str):
         if isinstance(document, str):
             img = Image.open(document).convert("RGB")
             img_array = np.array(img).transpose(2, 0, 1)
-            document = torch.tensor(img_array)
+            document = torch.from_numpy(img_array)
         pixel_values = self.pre_processor(document, return_tensors="pt").pixel_values
 
         return {"decoder_input_ids": decoder_input_ids, "pixel_values": pixel_values}

src/transformers/data/data_collator.py

@@ -153,7 +153,7 @@ def torch_default_data_collator(features: List[InputDataClass]) -> Dict[str, Any
             if isinstance(v, torch.Tensor):
                 batch[k] = torch.stack([f[k] for f in features])
             elif isinstance(v, np.ndarray):
-                batch[k] = torch.tensor(np.stack([f[k] for f in features]))
+                batch[k] = torch.from_numpy(np.stack([f[k] for f in features]))
             else:
                 batch[k] = torch.tensor([f[k] for f in features])
 

src/transformers/feature_extraction_utils.py

@@ -146,7 +146,10 @@ def as_tensor(value):
                         and isinstance(value[0][0], np.ndarray)
                     ):
                         value = np.array(value)
-                return torch.tensor(value)
+                if isinstance(value, np.ndarray):
+                    return torch.from_numpy(value)
+                else:
+                    return torch.tensor(value)
 
             is_tensor = torch.is_tensor
         elif tensor_type == TensorType.JAX:

src/transformers/image_utils.py

@@ -579,9 +579,15 @@ def normalize(self, image, mean, std, rescale=False):
             import torch
 
             if not isinstance(mean, torch.Tensor):
-                mean = torch.tensor(mean)
+                if isinstance(mean, np.ndarray):
+                    mean = torch.from_numpy(mean)
+                else:
+                    mean = torch.tensor(mean)
             if not isinstance(std, torch.Tensor):
-                std = torch.tensor(std)
+                if isinstance(std, np.ndarray):
+                    std = torch.from_numpy(std)
+                else:
+                    std = torch.tensor(std)
 
         if image.ndim == 3 and image.shape[0] in [1, 3]:
             return (image - mean[:, None, None]) / std[:, None, None]

src/transformers/tokenization_utils_base.py

@@ -721,7 +721,7 @@ def convert_to_tensors(
 
             def as_tensor(value, dtype=None):
                 if isinstance(value, list) and isinstance(value[0], np.ndarray):
-                    return torch.tensor(np.array(value))
+                    return torch.from_numpy(np.array(value))
                 return torch.tensor(value)
 
         elif tensor_type == TensorType.JAX: