update app

app.py

@@ -136,6 +136,15 @@ model_y = Qwen3VLForConditionalGeneration.from_pretrained(
     torch_dtype=torch.float16
 ).to(device).eval()
 
+# Load Qwen3-VL-4B-Thinking
+MODEL_ID_T = "Qwen/Qwen3-VL-4B-Thinking"
+processor_t = AutoProcessor.from_pretrained(MODEL_ID_T, trust_remote_code=True)
+model_t = Qwen3VLForConditionalGeneration.from_pretrained(
+    MODEL_ID_T,
+    trust_remote_code=True,
+    torch_dtype=torch.float16
+).to(device).eval()
+
 def downsample_video(video_path):
     """
     Downsamples the video to evenly spaced frames.
@@ -176,6 +185,8 @@ def generate_image(model_name: str, text: str, image: Image.Image,
         processor, model = processor_q, model_q
     elif model_name == "Qwen3-VL-8B-Instruct":
         processor, model = processor_y, model_y
+    elif model_name == "Qwen3-VL-4B-Thinking":
+        processor, model = processor_t, model_t
     else:
         yield "Invalid model selected.", "Invalid model selected."
         return
@@ -219,6 +230,8 @@ def generate_video(model_name: str, text: str, video_path: str,
         processor, model = processor_q, model_q
     elif model_name == "Qwen3-VL-8B-Instruct":
         processor, model = processor_y, model_y
+    elif model_name == "Qwen3-VL-4B-Thinking":
+        processor, model = processor_t, model_t
     else:
         yield "Invalid model selected.", "Invalid model selected."
         return
@@ -315,7 +328,7 @@ with gr.Blocks(css=css, theme=steel_blue_theme) as demo:
                 markdown_output = gr.Markdown()
                 
             model_choice = gr.Radio(
-                choices=["Qwen3-VL-4B-Instruct", "Qwen3-VL-8B-Instruct", "Qwen2.5-VL-3B-Instruct", "Qwen2.5-VL-7B-Instruct"],
+                choices=["Qwen3-VL-4B-Instruct", "Qwen3-VL-8B-Instruct", "Qwen3-VL-4B-Thinking", "Qwen2.5-VL-3B-Instruct", "Qwen2.5-VL-7B-Instruct"],
                 label="Select Model",
                 value="Qwen3-VL-4B-Instruct"
             )