fix: add do_sample_frames flag to processor_kwargs to prevent metadata errors

app.py

@@ -115,13 +115,13 @@ def predict(message: str, file: FileData = None, downsample_mode: str = "16x") -
             inputs = processor.apply_chat_template(
                 messages, tokenize=True, add_generation_prompt=True,
                 return_dict=True, return_tensors="pt",
-                # Wrap vision kwargs in processor_kwargs for newer transformers versions
                 processor_kwargs={
                     "downsample_mode": downsample_mode,
                     "max_num_frames": 64,
                     "stack_frames": 1,
                     "max_slice_nums": 1,
                     "use_image_id": False,
+                    "do_sample_frames": False, # Fix: Avoid requiring metadata since we already sampled
                 }
             ).to(model.device)
         else:
@@ -138,7 +138,6 @@ def predict(message: str, file: FileData = None, downsample_mode: str = "16x") -
             inputs = processor.apply_chat_template(
                 messages, tokenize=True, add_generation_prompt=True,
                 return_dict=True, return_tensors="pt",
-                # Wrap vision kwargs in processor_kwargs
                 processor_kwargs={
                     "downsample_mode": downsample_mode,
                     "max_slice_nums": 9,