Update app.py

app.py

@@ -56,6 +56,15 @@ model_q = Qwen2_5_VLForConditionalGeneration.from_pretrained(
     torch_dtype=torch.float16
 ).to(device).eval()
 
+# Load YannQi/X-VL-4B
+MODEL_ID_I = "YannQi/X-VL-4B"
+processor_i= AutoProcessor.from_pretrained(MODEL_ID_I, trust_remote_code=True)
+model_i = AutoModel.from_pretrained(
+    MODEL_ID_I,
+    trust_remote_code=True,
+    torch_dtype=torch.float16
+).to(device).eval()
+
 def downsample_video(video_path):
     """
     Downsamples the video to evenly spaced frames.
@@ -97,6 +106,9 @@ def generate_image(model_name: str, text: str, image: Image.Image,
     elif model_name == "Qwen2.5-VL-7B-Abliterated-Caption-it":
         processor = processor_q
         model = model_q
+    elif model_name == "X-VL-4B":
+        processor = processor_i
+        model = model_i
     else:
         yield "Invalid model selected.", "Invalid model selected."
         return
@@ -151,6 +163,9 @@ def generate_video(model_name: str, text: str, video_path: str,
     elif model_name == "Qwen2.5-VL-7B-Abliterated-Caption-it":
         processor = processor_q
         model = model_q
+    elif model_name == "X-VL-4B":
+        processor = processor_i
+        model = model_i
     else:
         yield "Invalid model selected.", "Invalid model selected."
         return
@@ -262,7 +277,7 @@ with gr.Blocks(css=css, theme="bethecloud/storj_theme") as demo:
                     markdown_output = gr.Markdown()
                 
             model_choice = gr.Radio(
-                choices=["Qwen2.5-VL-7B-Instruct", "Qwen2.5-VL-3B-Instruct", "Qwen2.5-VL-7B-Abliterated-Caption-it"],
+                choices=["Qwen2.5-VL-7B-Instruct", "Qwen2.5-VL-3B-Instruct", "X-VL-4B", "Qwen2.5-VL-7B-Abliterated-Caption-it"],
                 label="Select Model",
                 value="Qwen2.5-VL-7B-Instruct"
             )
@@ -270,6 +285,7 @@ with gr.Blocks(css=css, theme="bethecloud/storj_theme") as demo:
             gr.Markdown("> [Qwen2.5-VL-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct): The Qwen2.5-VL-7B-Instruct model is a multimodal AI model developed by Alibaba Cloud that excels at understanding both text and images. It's a Vision-Language Model (VLM) designed to handle various visual understanding tasks, including image understanding, video analysis, and even multilingual support.")
             gr.Markdown("> [Qwen2.5-VL-3B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-3B-Instruct): Qwen2.5-VL-3B-Instruct is an instruction-tuned vision-language model from Alibaba Cloud, built upon the Qwen2-VL series. It excels at understanding and generating text related to both visual and textual inputs, making it capable of tasks like image captioning, visual question answering, and object localization. The model also supports long video understanding and structured data extraction")
             gr.Markdown("> [Qwen2.5-VL-7B-Abliterated-Caption-it](prithivMLmods/Qwen2.5-VL-7B-Abliterated-Caption-it): Qwen2.5-VL-7B-Abliterated-Caption-it is a fine-tuned version of Qwen2.5-VL-7B-Instruct, optimized for Abliterated Captioning / Uncensored Captioning. This model excels at generating detailed, context-rich, and high-fidelity captions across diverse image categories and variational aspect ratios, offering robust visual understanding without filtering or censorship.")
+            gr.Markdown("> [X-VL-4B](huggingface.co/YannQi/X-VL-4B): X-VL-4B, a multimodal large language model designed to achieve adaptive multimodal reasoning—dynamically choosing between step-by-step thinking and direct response generation based on task complexity. This capability enables X-VL-4B to deliver high-quality responses while significantly improving inference efficiency and reducing computational costs")
             gr.Markdown(">⚠️note: all the models in space are not guaranteed to perform well in video inference use cases.")
             
     image_submit.click(