Ignore type checks for torch.cuda.stream

Add `# type: ignore` to two usages of `torch.cuda.stream` (in trt_base.py and vila_processor.py) to suppress static type-checker complaints about the CUDA stream context manager. This is a no-op at runtime and only affects linting/type-checking to keep CI clean.

Author: openminddev

src/om1_vlm/anonymizationSys/face_recog_stream/trt_base.py

@@ -110,7 +110,7 @@ def _to_gpu(self, host_np: np.ndarray) -> torch.Tensor:
 
         Use ``tensor.data_ptr()`` to get the raw pointer for TRT.
         """
-        with torch.cuda.stream(self._torch_stream):
+        with torch.cuda.stream(self._torch_stream):  # type: ignore
             return torch.from_numpy(host_np).cuda(non_blocking=True)
 
     def _empty_gpu(self, shape, dtype=torch.float32) -> torch.Tensor:

src/om1_vlm/vila_vlm/vila_processor.py

@@ -211,7 +211,7 @@ def generate_with_images(self, images: List[bytes], prompt: str):
             with self.model_lock:
                 # Set the current CUDA stream
                 if self.cuda_stream:
-                    with torch.cuda.stream(self.cuda_stream):
+                    with torch.cuda.stream(self.cuda_stream):  # type: ignore
                         # Add explicit CUDA synchronization before processing
                         if torch.cuda.is_available():
                             torch.cuda.synchronize()