still use sample

src/video_processor/processor.py

@@ -104,7 +104,7 @@ Be specific about all visual elements to enable searching later."""}
         with torch.inference_mode():
             outputs = self.model.generate(
                 **inputs,
-                do_sample=False,
+                do_sample=True,
                 temperature=0.7,
                 max_new_tokens=256,
             )