Implement custom tensor.isin

TTS/tts/layers/xtts/stream_generator.py

@@ -23,6 +23,20 @@ from transformers import (
 )
 from transformers.generation.utils import GenerateOutput, SampleOutput, logger
 
+def custom_isin(elements, test_elements):
+    # Flatten the tensors
+    elements_flat = elements.view(-1)
+    test_elements_flat = test_elements.view(-1)
+
+    # Create a mask tensor
+    mask = torch.zeros_like(elements_flat, dtype=torch.bool)
+
+    # Compare each element
+    for test_element in test_elements_flat:
+        mask |= (elements_flat == test_element)
+
+    # Reshape the mask to the original elements shape
+    return mask.view(elements.shape)
 
 def setup_seed(seed):
     if seed == -1:
@@ -202,10 +216,10 @@ class NewGenerationMixin(GenerationMixin):
                 default_attention_mask = torch.ones(inputs_tensor.shape[:2], dtype=torch.long, device=inputs_tensor.device)
 
                 is_pad_token_in_inputs = (pad_token_tensor is not None) and (
-                    torch.isin(elements=inputs_tensor, test_elements=pad_token_tensor).any()
+                    custom_isin(elements=inputs_tensor, test_elements=pad_token_tensor).any()
                 )
                 is_pad_token_not_equal_to_eos_token_id = (eos_token_tensor is None) or ~(
-                    torch.isin(elements=eos_token_tensor, test_elements=pad_token_tensor).any()
+                    custom_isin(elements=eos_token_tensor, test_elements=pad_token_tensor).any()
                 )
                 can_infer_attention_mask = is_pad_token_in_inputs * is_pad_token_not_equal_to_eos_token_id
                 attention_mask_from_padding = inputs_tensor.ne(pad_token_tensor).long()