Update app.py

app.py

@@ -70,6 +70,15 @@ model_f = Qwen2_5_VLForConditionalGeneration.from_pretrained(
     torch_dtype=torch.float16
 ).to(device).eval()
 
+# Load DeepEyes-7B
+MODEL_ID_Y = "ChenShawn/DeepEyes-7B"
+processor_y = AutoProcessor.from_pretrained(MODEL_ID_Y, trust_remote_code=True)
+model_y = Qwen2_5_VLForConditionalGeneration.from_pretrained(
+    MODEL_ID_Y,
+    trust_remote_code=True,
+    torch_dtype=torch.float16
+).to(device).eval()
+
 def downsample_video(video_path):
     """
     Downsample a video to evenly spaced frames, returning each as a PIL image with its timestamp.
@@ -115,6 +124,9 @@ def generate_image(model_name: str, text: str, image: Image.Image,
     elif model_name == "OVR-7B-RL":
         processor = processor_f
         model = model_f
+    elif model_name == "DeepEyes-7B-Thinking":
+        processor = processor_y
+        model = model_y
     else:
         yield "Invalid model selected.", "Invalid model selected."
         return
@@ -174,6 +186,9 @@ def generate_video(model_name: str, text: str, video_path: str,
     elif model_name == "OVR-7B-RL":
         processor = processor_f
         model = model_f
+    elif model_name == "DeepEyes-7B-Thinking":
+        processor = processor_y
+        model = model_y
     else:
         yield "Invalid model selected.", "Invalid model selected."
         return
@@ -289,7 +304,7 @@ with gr.Blocks(css=css, theme="bethecloud/storj_theme") as demo:
                     markdown_output = gr.Markdown(label="(Result.md)")
 
             model_choice = gr.Radio(
-                choices=["Camel-Doc-OCR-062825", "GLM-4.1V-9B-Thinking", "Megalodon-OCR-Sync-0713", "MonkeyOCR-pro-1.2B", "OVR-7B-RL"],
+                choices=["Camel-Doc-OCR-062825", "GLM-4.1V-9B-Thinking", "Megalodon-OCR-Sync-0713", "MonkeyOCR-pro-1.2B", "DeepEyes-7B-Thinking", "OVR-7B-RL"],
                 label="Select Model",
                 value="Camel-Doc-OCR-062825"
             )