Update app.py

app.py

@@ -55,12 +55,16 @@ model_o = Qwen2_5_VLForConditionalGeneration.from_pretrained(
     MODEL_ID_O, trust_remote_code=True,
     torch_dtype=torch.float16).to(device).eval()
 
-# Load VLM-R1-Qwen2.5VL-3B-Math-0305
-MODEL_ID_W = "omlab/VLM-R1-Qwen2.5VL-3B-Math-0305"
-processor_w = AutoProcessor.from_pretrained(MODEL_ID_W, trust_remote_code=True)
+#-----------------------------subfolder-----------------------------#
+# Load MonkeyOCR-3B-0709
+MODEL_ID_W = "echo840/MonkeyOCR-3B-0709"
+SUBFOLDER = "Recognition"
+processor_w = AutoProcessor.from_pretrained(MODEL_ID_W, trust_remote_code=True, subfolder=SUBFOLDER)
 model_w = Qwen2_5_VLForConditionalGeneration.from_pretrained(
     MODEL_ID_W, trust_remote_code=True,
+    subfolder=SUBFOLDER,
     torch_dtype=torch.float16).to(device).eval()
+#-----------------------------subfolder-----------------------------#
 
 # Function to downsample video frames
 def downsample_video(video_path):
@@ -109,7 +113,7 @@ def generate_image(model_name: str,
     elif model_name == "R1-Onevision-7B":
         processor = processor_t
         model = model_t
-    elif model_name == "VLM-R1-Qwen2.5VL-3B-Math-0305":
+    elif model_name == "MonkeyOCR-3B-0709":
         processor = processor_w
         model = model_w
     else:
@@ -176,7 +180,7 @@ def generate_video(model_name: str,
     elif model_name == "R1-Onevision-7B":
         processor = processor_t
         model = model_t
-    elif model_name == "VLM-R1-Qwen2.5VL-3B-Math-0305":
+    elif model_name == "MonkeyOCR-3B-0709":
         processor = processor_w
         model = model_w
     else:
@@ -326,8 +330,7 @@ with gr.Blocks(css=css, theme="bethecloud/storj_theme") as demo:
                 #download_btn = gr.Button("Download Result.md")
 
             model_choice = gr.Radio(choices=[
-                "Vision-Matters-7B-Math", "ViGaL-7B", "Visionary-R1",
-                "R1-Onevision-7B", "VLM-R1-Qwen2.5VL-3B-Math-0305"
+                "Vision-Matters-7B-Math", "MonkeyOCR-3B-0709", "ViGaL-7B", "Visionary-R1", "R1-Onevision-7B"
             ],
                                     label="Select Model",
                                     value="Vision-Matters-7B-Math")
@@ -337,7 +340,7 @@ with gr.Blocks(css=css, theme="bethecloud/storj_theme") as demo:
             gr.Markdown("> [ViGaL 7B](https://huggingface.co/yunfeixie/ViGaL-7B): vigal-7b shows that training a 7b mllm on simple games like snake using reinforcement learning boosts performance on benchmarks like mathvista and mmmu without needing worked solutions or diagrams indicating transferable reasoning skills.")
             gr.Markdown("> [Visionary-R1](https://huggingface.co/maifoundations/Visionary-R1): visionary-r1 is a novel framework for training visual language models (vlms) to perform robust visual reasoning using reinforcement learning (rl). unlike traditional approaches that rely heavily on (sft) or (cot) annotations, visionary-r1 leverages only visual question-answer pairs and rl, making the process more scalable and accessible.")
             gr.Markdown("> [R1-Onevision-7B](https://huggingface.co/Fancy-MLLM/R1-Onevision-7B): r1-onevision model enhances vision-language understanding and reasoning capabilities, making it suitable for various tasks such as visual reasoning and image understanding. with its robust ability to perform multimodal reasoning, r1-onevision emerges as a powerful ai assistant capable of addressing different domains.")
-            gr.Markdown("> [VLM-R1-Qwen2.5VL-3B-Math-0305](https://huggingface.co/omlab/VLM-R1-Qwen2.5VL-3B-Math-0305): vlm-r1 is a framework designed to enhance the reasoning and generalization capabilities of vision-language models (vlms) using a reinforcement learning (rl) approach inspired by the r1 methodology originally developed for large language models.")
+            gr.Markdown("> [MonkeyOCR-3B-0709](https://huggingface.co/omlab/VLM-R1-Qwen2.5VL-3B-Math-0305): vlm-r1 is a framework designed to enhance the reasoning and generalization capabilities of vision-language models (vlms) using a reinforcement learning (rl) approach inspired by the r1 methodology originally developed for large language models.")
             gr.Markdown(">⚠️note: all the models in space are not guaranteed to perform well in video inference use cases.")  
 
     # Define the submit button actions