Update app.py

app.py

@@ -20,7 +20,7 @@ import fitz
 
 from transformers import (
     Qwen2_5_VLForConditionalGeneration,
-    Blip2ForConditionalGeneration,
+    Qwen2VLForConditionalGeneration,
     AutoModelForCausalLM,
     AutoModelForVision2Seq,
     AutoModelForImageTextToText,
@@ -79,18 +79,24 @@ model_g = Qwen2_5_VLForConditionalGeneration.from_pretrained(
     MODEL_ID_G, trust_remote_code=True, subfolder=SUBFOLDER, torch_dtype=torch.float16
 ).to(device).eval()
 
-MODEL_ID_I = "Salesforce/blip2-opt-2.7b"
+MODEL_ID_I = "UCSC-VLAA/VLAA-Thinker-Qwen2VL-2B"
 processor_i = AutoProcessor.from_pretrained(MODEL_ID_I, trust_remote_code=True)
-model_i = Blip2ForConditionalGeneration.from_pretrained(
+model_i = Qwen2VLForConditionalGeneration.from_pretrained(
     MODEL_ID_I, trust_remote_code=True, torch_dtype=torch.float16
 ).to(device).eval()
 
-MODEL_ID_A = "facebook/Perception-LM-1B"
+MODEL_ID_A = "nanonets/Nanonets-OCR-s"
 processor_a = AutoProcessor.from_pretrained(MODEL_ID_A, trust_remote_code=True)
-model_a = AutoModelForImageTextToText.from_pretrained(
+model_a = Qwen2_5_VLForConditionalGeneration.from_pretrained(
     MODEL_ID_A, trust_remote_code=True, torch_dtype=torch.float16
 ).to(device).eval()
 
+MODEL_ID_X = "prithivMLmods/Megalodon-OCR-Sync-0713"
+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()
+
 # --- PDF Generation and Preview Utility Function ---
 def generate_and_preview_pdf(image: Image.Image, text_content: str, font_size: int, line_spacing: float, alignment: str, image_size: str):
     """
@@ -185,8 +191,9 @@ def process_document_stream(
     elif model_name == "LFM2-VL-1.6B": processor, model = processor_t, model_t
     elif model_name == "SmolVLM-Instruct-250M": processor, model = processor_c, model_c
     elif model_name == "MonkeyOCR-pro-1.2B": processor, model = processor_g, model_g
-    elif model_name == "blip2-opt-2.7b": processor, model = processor_i, model_i
-    elif model_name == "Perception-LM-1B": processor, model = processor_a, model_a
+    elif model_name == "VLAA-Thinker-Qwen2VL-2B": processor, model = processor_i, model_i
+    elif model_name == "Nanonets-OCR-s": processor, model = processor_a, model_a
+    elif model_name == "Megalodon-OCR-Sync-0713": processor, model = processor_x, model_x
     else:
         yield "Invalid model selected.", ""
         return
@@ -243,7 +250,8 @@ def create_gradio_interface():
             # Left Column (Inputs)
             with gr.Column(scale=1):
                 model_choice = gr.Dropdown(
-                    choices=["LFM2-VL-1.6B", "LFM2-VL-450M", "SmolVLM-Instruct-250M", "blip2-opt-2.7b", "MonkeyOCR-pro-1.2B", "Perception-LM-1B"],
+                    choices=["LFM2-VL-1.6B", "LFM2-VL-450M", "SmolVLM-Instruct-250M", "VLAA-Thinker-Qwen2VL-2B", "MonkeyOCR-pro-1.2B",
+                             "Nanonets-OCR-s", "Megalodon-OCR-Sync-0713"],
                     label="Select Model", value="LFM2-VL-1.6B"
                 )
                 prompt_input = gr.Textbox(label="Query Input", placeholder="✦︎ Enter your query")