Update app.py

app.py

@@ -16,6 +16,7 @@ import cv2
 from transformers import (
     Qwen2VLForConditionalGeneration,
     Qwen2_5_VLForConditionalGeneration,
+    VisionEncoderDecoderModel,
     AutoModelForVision2Seq,
     AutoProcessor,
     TextIteratorStreamer,
@@ -44,6 +45,16 @@ model_m = Qwen2_5_VLForConditionalGeneration.from_pretrained(
     torch_dtype=torch.float16
 ).to(device).eval()
 
+# Load ByteDance's Dolphin
+MODEL_ID_K = "ByteDance/Dolphin"
+processor_k = AutoProcessor.from_pretrained(MODEL_ID_K, trust_remote_code=True)
+model_k = VisionEncoderDecoderModel.from_pretrained(
+    MODEL_ID_K,
+    trust_remote_code=True,
+    torch_dtype=torch.float16
+).to(device).eval()
+
+#------------------------------------------------#
 # Load SmolDocling-256M-preview
 MODEL_ID_X = "ds4sd/SmolDocling-256M-preview"
 processor_x = AutoProcessor.from_pretrained(MODEL_ID_X, trust_remote_code=True)
@@ -52,6 +63,7 @@ model_x = AutoModelForVision2Seq.from_pretrained(
     trust_remote_code=True,
     torch_dtype=torch.float16
 ).to(device).eval()
+#------------------------------------------------#
 
 # Load MonkeyOCR
 MODEL_ID_G = "echo840/MonkeyOCR"
@@ -132,6 +144,9 @@ def generate_image(model_name: str, text: str, image: Image.Image,
     elif model_name == "SmolDocling-256M-preview":
         processor = processor_x
         model = model_x
+    elif model_name == "ByteDance-s-Dolphin":
+        processor = processor_k
+        model = model_k
     else:
         yield "Invalid model selected."
         return
@@ -216,6 +231,9 @@ def generate_video(model_name: str, text: str, video_path: str,
     elif model_name == "SmolDocling-256M-preview":
         processor = processor_x
         model = model_x
+    elif model_name == "ByteDance-s-Dolphin":
+        processor = processor_k
+        model = model_k
     else:
         yield "Invalid model selected."
         return
@@ -336,7 +354,7 @@ with gr.Blocks(css=css, theme="bethecloud/storj_theme") as demo:
         with gr.Column():
             output = gr.Textbox(label="Output", interactive=False, lines=3, scale=2)
             model_choice = gr.Radio(
-                choices=["Nanonets-OCR-s", "SmolDocling-256M-preview", "MonkeyOCR-Recognition"],
+                choices=["Nanonets-OCR-s", "SmolDocling-256M-preview", "MonkeyOCR-Recognition", "ByteDance-s-Dolphin"],
                 label="Select Model",
                 value="Nanonets-OCR-s"
             )