Update app.py

app.py

@@ -15,7 +15,7 @@ import cv2
 
 from transformers import (
     Qwen2_5_VLForConditionalGeneration,
-    LlavaOnevisionForConditionalGeneration,
+    AutoModelForCausalLM,
     AutoProcessor,
     TextIteratorStreamer,
 )
@@ -35,25 +35,33 @@ model_m = Qwen2_5_VLForConditionalGeneration.from_pretrained(
     MODEL_ID_M, trust_remote_code=True,
     torch_dtype=torch.float16).to(device).eval()
 
-# Load ViGaL-7B
-MODEL_ID_X = "yunfeixie/ViGaL-7B"
+# Load OCRFlux-3B
+MODEL_ID_X = "ChatDOC/OCRFlux-3B"
 processor_x = AutoProcessor.from_pretrained(MODEL_ID_X, trust_remote_code=True)
 model_x = Qwen2_5_VLForConditionalGeneration.from_pretrained(
     MODEL_ID_X, trust_remote_code=True,
     torch_dtype=torch.float16).to(device).eval()
 
-# Load prithivMLmods/WR30a-Deep-7B-0711
-MODEL_ID_T = "NCSOFT/VARCO-VISION-2.0-1.7B-OCR"
+# Load A.X-4.0-VL-Light
+MODEL_ID_T = "skt/A.X-4.0-VL-Light"
 processor_t = AutoProcessor.from_pretrained(MODEL_ID_T, trust_remote_code=True)
-model_t = LlavaOnevisionForConditionalGeneration.from_pretrained(
+model_t = AutoModelForCausalLM.from_pretrained(
     MODEL_ID_T, trust_remote_code=True,
     torch_dtype=torch.float16).to(device).eval()
 
-# Load Visionary-R1
-MODEL_ID_O = "maifoundations/Visionary-R1"
-processor_o = AutoProcessor.from_pretrained(MODEL_ID_O, trust_remote_code=True)
+# Load MonkeyOCR-pro-1.2B
+MODEL_ID_O = "echo840/MonkeyOCR-pro-1.2B"
+SUBFOLDER = "Recognition"
+processor_o = AutoProcessor.from_pretrained(MODEL_ID_O, trust_remote_code=True, subfolder=SUBFOLDER)
 model_o = Qwen2_5_VLForConditionalGeneration.from_pretrained(
-    MODEL_ID_O, trust_remote_code=True,
+    MODEL_ID_O, trust_remote_code=True, subfolder=SUBFOLDER, 
+    torch_dtype=torch.float16).to(device).eval()
+
+# Load OpenVLThinker-7B-v1.2
+MODEL_ID_A = "ydeng9/OpenVLThinker-7B-v1.2"
+processor_a = AutoProcessor.from_pretrained(MODEL_ID_A, trust_remote_code=True)
+model_a = Qwen2_5_VLForConditionalGeneration.from_pretrained(
+    MODEL_ID_A, trust_remote_code=True,
     torch_dtype=torch.float16).to(device).eval()
 
 # Function to downsample video frames
@@ -94,15 +102,18 @@ def generate_image(model_name: str,
     if model_name == "Camel-Doc-OCR-080125(v2)":
         processor = processor_m
         model = model_m
-    elif model_name == "ViGaL-7B":
+    elif model_name == "OCRFlux-3B":
         processor = processor_x
         model = model_x
-    elif model_name == "Visionary-R1-3B":
+    elif model_name == "A.X-4.0-VL-Light":
         processor = processor_o
         model = model_o
-    elif model_name == "Varco-Vision-2.0-OCR":
+    elif model_name == "MonkeyOCR-pro-1.2B":
         processor = processor_t
         model = model_t
+    elif model_name == "OpenVLThinker-7B":
+        processor = processor_a
+        model = model_a
     else:
         yield "Invalid model selected.", "Invalid model selected."
         return
@@ -158,15 +169,18 @@ def generate_video(model_name: str,
     if model_name == "Camel-Doc-OCR-080125(v2)":
         processor = processor_m
         model = model_m
-    elif model_name == "ViGaL-7B":
+    elif model_name == "OCRFlux-3B":
         processor = processor_x
         model = model_x
-    elif model_name == "Visionary-R1-3B":
+    elif model_name == "A.X-4.0-VL-Light":
         processor = processor_o
         model = model_o
-    elif model_name == "Varco-Vision-2.0-OCR":
+    elif model_name == "MonkeyOCR-pro-1.2B":
         processor = processor_t
         model = model_t
+    elif model_name == "OpenVLThinker-7B":
+        processor = processor_a
+        model = model_a
     else:
         yield "Invalid model selected.", "Invalid model selected."
         return
@@ -249,7 +263,7 @@ css = """
 # Create the Gradio Interface
 with gr.Blocks(css=css, theme="bethecloud/storj_theme") as demo:
     gr.Markdown(
-        "# **[Multimodal VLM OCR3](https://huggingface.co/collections/prithivMLmods/multimodal-implementations-67c9982ea04b39f0608badb0)**"
+        "# **[Multimodal OCR Outpost](https://huggingface.co/collections/prithivMLmods/multimodal-implementations-67c9982ea04b39f0608badb0)**"
     )
     with gr.Row():
         with gr.Column():
@@ -312,20 +326,12 @@ with gr.Blocks(css=css, theme="bethecloud/storj_theme") as demo:
                 #download_btn = gr.Button("Download Result.md")
 
             model_choice = gr.Radio(choices=[
-                 "Camel-Doc-OCR-080125(v2)", "Varco-Vision-2.0-OCR",
-                 "ViGaL-7B", "Visionary-R1-3B"
+                 "Camel-Doc-OCR-080125(v2)", "OCRFlux-3B", "OpenVLThinker-7B",
+                 "A.X-4.0-VL-Light", "MonkeyOCR-pro-1.2B"
             ],
                                     label="Select Model",
                                     value="Camel-Doc-OCR-080125(v2)")
 
-            gr.Markdown("**Model Info 💻** | [Report Bug](https://huggingface.co/spaces/prithivMLmods/Multimodal-VLMs-5x/discussions)")         
-            gr.Markdown("> [Camel-Doc-OCR-080125(v2)](https://huggingface.co/prithivMLmods/WR30a-Deep-7B-0711): the camel-doc-ocr-080125 model is a fine-tuned version of qwen2.5-vl-7b-instruct, optimized for document retrieval, content extraction, and analysis recognition. built on top of the qwen2.5-vl architecture, this model enhances document comprehension capabilities.")
-            gr.Markdown("> [MonkeyOCR-pro-1.2B](https://huggingface.co/echo840/MonkeyOCR-pro-1.2B): MonkeyOCR adopts a structure-recognition-relation (SRR) triplet paradigm, which simplifies the multi-tool pipeline of modular approaches while avoiding the inefficiency of using large multimodal models for full-page document processing.")
-            gr.Markdown("> [Vision Matters 7B](https://huggingface.co/Yuting6/Vision-Matters-7B): vision-matters is a simple visual perturbation framework that can be easily integrated into existing post-training pipelines including sft, dpo, and grpo. our findings highlight the critical role of visual perturbation: better reasoning begins with better seeing.")
-            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(">⚠️note: all the models in space are not guaranteed to perform well in video inference use cases.")  
-
     # Define the submit button actions
     image_submit.click(fn=generate_image,
                        inputs=[