Update app.py

app.py

@@ -15,7 +15,6 @@ import cv2
 
 from transformers import (
     Qwen2_5_VLForConditionalGeneration,
-    Qwen2VLForConditionalGeneration,
     AutoProcessor,
     AutoTokenizer,
     TextIteratorStreamer,
@@ -34,7 +33,7 @@ MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
 device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
 
 # Load typhoon
-MODEL_ID_M = "scb10x/typhoon-ocr-7b"
+MODEL_ID_M = "Qwen/Qwen2.5-VL-3B-Instruct"
 processor_m = AutoProcessor.from_pretrained(MODEL_ID_M, trust_remote_code=True)
 model_m = Qwen2_5_VLForConditionalGeneration.from_pretrained(
     MODEL_ID_M,
@@ -42,17 +41,8 @@ model_m = Qwen2_5_VLForConditionalGeneration.from_pretrained(
     torch_dtype=torch.float16
 ).to(device).eval()
 
-# Load DocScope
-MODEL_ID_X = "prithivMLmods/coreOCR-7B-050325-preview"
-processor_x = AutoProcessor.from_pretrained(MODEL_ID_X, trust_remote_code=True)
-model_x = Qwen2VLForConditionalGeneration.from_pretrained(
-    MODEL_ID_X,
-    trust_remote_code=True,
-    torch_dtype=torch.float16
-).to(device).eval()
-
 # Load Space Thinker
-MODEL_ID_Z = "remyxai/SpaceThinker-Qwen2.5VL-3B"
+MODEL_ID_Z = "One-RL-to-See-Them-All/Orsta-32B-0326"
 processor_z = AutoProcessor.from_pretrained(MODEL_ID_Z, trust_remote_code=True)
 model_z = Qwen2_5_VLForConditionalGeneration.from_pretrained(
     MODEL_ID_Z,
@@ -60,15 +50,6 @@ model_z = Qwen2_5_VLForConditionalGeneration.from_pretrained(
     torch_dtype=torch.float16
 ).to(device).eval()
 
-# Load Qwen2-VL-7B-Instruct
-MODEL_ID_T = "Qwen/Qwen2-VL-7B-Instruct"
-processor_t = AutoTokenizer.from_pretrained(MODEL_ID_X, trust_remote_code=True)
-model_t = Qwen2VLForConditionalGeneration.from_pretrained(
-    MODEL_ID_T,
-    trust_remote_code=True,
-    torch_dtype=torch.float16
-).to(device).eval()
-
 
 
 def downsample_video(video_path):
@@ -102,18 +83,12 @@ def generate_image(model_name: str, text: str, image: Image.Image,
     """
     Generates responses using the selected model for image input.
     """
-    if model_name == "typhoon-ocr-7b":
+    if model_name == "Qwen2.5-VL-3B":
         processor = processor_m
         model = model_m
-    elif model_name == "coreOCR-7B-050325-preview":
-        processor = processor_x
-        model = model_x
-    elif model_name == "SpaceThinker-Qwen2.5VL-3B":
+    elif model_name == "Orsta-32B-0326":
         processor = processor_z
         model = model_z
-    elif model_name == "Qwen2-VL-7B-Instruct":
-        processor = processor_t
-        model = model_t
     else:
         yield "Invalid model selected."
         return
@@ -158,18 +133,12 @@ def generate_video(model_name: str, text: str, video_path: str,
     """
     Generates responses using the selected model for video input.
     """
-    if model_name == "typhoon-ocr-7b":
+    if model_name == "Qwen2.5-VL-3B":
         processor = processor_m
         model = model_m
-    elif model_name == "coreOCR-7B-050325-preview":
-        processor = processor_x
-        model = model_x
-    elif model_name == "SpaceThinker-Qwen2.5VL-3B":
+    elif model_name == "Orsta-32B-0326":
         processor = processor_z
         model = model_z
-    elif model_name == "Qwen2-VL-7B-Instruct":
-        processor = processor_t
-        model = model_t
     else:
         yield "Invalid model selected."
         return
@@ -270,9 +239,9 @@ with gr.Blocks(css=css, theme="bethecloud/storj_theme") as demo:
         with gr.Column():
             output = gr.Textbox(label="Output", interactive=False, lines=2, scale=2)
             model_choice = gr.Radio(
-                choices=["coreOCR-7B-050325-preview", "Qwen2-VL-7B-Instruct", "SpaceThinker-Qwen2.5VL-3B", "typhoon-ocr-7b"],
+                choices=["Qwen2.5-VL-3B", "Orsta-32B-0326"],
                 label="Select Model",
-                value="Qwen2-VL-7B-Instruct"
+                value="Orsta-32B-0326"
             )
             
     image_submit.click(
@@ -287,4 +256,4 @@ with gr.Blocks(css=css, theme="bethecloud/storj_theme") as demo:
     )
 
 if __name__ == "__main__":
-    demo.queue(max_size=40).launch(share=True, mcp_server=True, ssr_mode=False, show_error=True)
+    demo.queue(max_size=30).launch(share=True, mcp_server=True, ssr_mode=False, show_error=True)