Add post_process_image_text_to_text to idefics3, mllama, pixtral proc…

…essor

src/transformers/models/idefics3/processing_idefics3.py

@@ -331,6 +331,20 @@ def decode(self, *args, **kwargs):
         decode_output = self.tokenizer.decode(*args, **kwargs)
         return self._regex_to_remove_extra_special_tokens.sub("<image>", decode_output)
 
+    def post_process_image_text_to_text(self, generated_outputs):
+        """
+        Post-process the output of the model to decode the text.
+
+        Args:
+            generated_outputs (`torch.Tensor` or `np.ndarray`):
+                The output of the model `generate` function. The output is expected to be a tensor of shape `(batch_size, sequence_length)`
+                or `(sequence_length,)`.
+
+        Returns:
+            `List[str]`: The decoded text.
+        """
+        return self.tokenizer.batch_decode(generated_outputs, skip_special_tokens=True)
+
     @property
     def model_input_names(self):
         tokenizer_input_names = self.tokenizer.model_input_names

src/transformers/models/mllama/processing_mllama.py

@@ -342,6 +342,22 @@ def decode(self, *args, **kwargs):
         """
         return self.tokenizer.decode(*args, **kwargs)
 
+    def post_process_image_text_to_text(self, generated_outputs):
+        """
+        Post-process the output of the model to decode the text.
+
+        Args:
+            generated_outputs (`torch.Tensor` or `np.ndarray`):
+                The output of the model `generate` function. The output is expected to be a tensor of shape `(batch_size, sequence_length)`
+                or `(sequence_length,)`.
+
+        Returns:
+            `List[str]`: The decoded text.
+        """
+        return self.tokenizer.batch_decode(
+            generated_outputs, skip_special_tokens=True, clean_up_tokenization_spaces=False
+        )
+
     @property
     def model_input_names(self):
         tokenizer_input_names = self.tokenizer.model_input_names

src/transformers/models/paligemma/processing_paligemma.py

@@ -100,7 +100,7 @@ def build_string_from_input(prompt, bos_token, image_seq_len, image_token, num_i
     if image_token in prompt:
         warnings.warn(
             f"The image token {image_token} is already present in the prompt. No need to manually add {image_token} in the prompt for this model."
-            f" Remove all {image_token} and adding ({image_token}) * image_seq_len at the start of the prompt."
+            f" Removing all {image_token} and adding ({image_token}) * image_seq_len * num_images at the start of the prompt."
         )
         prompt = prompt.replace(image_token, "")
     return f"{image_token * image_seq_len * num_images}{bos_token}{prompt}\n"

src/transformers/models/pixtral/processing_pixtral.py

@@ -274,6 +274,20 @@ def decode(self, *args, **kwargs):
         """
         return self.tokenizer.decode(*args, **kwargs)
 
+    def post_process_image_text_to_text(self, generated_outputs):
+        """
+        Post-process the output of the model to decode the text.
+
+        Args:
+            generated_outputs (`torch.Tensor` or `np.ndarray`):
+                The output of the model `generate` function. The output is expected to be a tensor of shape `(batch_size, sequence_length)`
+                or `(sequence_length,)`.
+
+        Returns:
+            `List[str]`: The decoded text.
+        """
+        return self.tokenizer.batch_decode(generated_outputs, skip_special_tokens=True)
+
     @property
     # Copied from transformers.models.clip.processing_clip.CLIPProcessor.model_input_names
     def model_input_names(self):