From f0110b7048469e0d5cbb697456f7b1cd80af37ee Mon Sep 17 00:00:00 2001 From: abrar Date: Wed, 25 Feb 2026 21:15:31 +0000 Subject: [PATCH] [Serve] fix video analysis release test Signed-off-by: abrar --- .../serve/tutorials/video-analysis/deployments/encoder.py | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) diff --git a/doc/source/serve/tutorials/video-analysis/deployments/encoder.py b/doc/source/serve/tutorials/video-analysis/deployments/encoder.py index 5815b4e12833..4306cc420e07 100644 --- a/doc/source/serve/tutorials/video-analysis/deployments/encoder.py +++ b/doc/source/serve/tutorials/video-analysis/deployments/encoder.py @@ -69,9 +69,8 @@ def encode_frames(self, frames: np.ndarray) -> np.ndarray: with torch.no_grad(): with torch.amp.autocast(device_type=self.device, enabled=self.device == "cuda"): outputs = self.model.get_image_features(**inputs) - - # L2 normalize on GPU (faster than CPU numpy) - frame_embeddings = torch.nn.functional.normalize(outputs, p=2, dim=1) + # get_image_features returns BaseModelOutputWithPooling; use pooler_output for embeddings + frame_embeddings = torch.nn.functional.normalize(outputs.pooler_output, p=2, dim=1) # Move to CPU and convert to numpy result = frame_embeddings.cpu().numpy().astype(np.float32)