Update app.py

app.py

@@ -138,10 +138,6 @@ model_m = Qwen2_5_VLForConditionalGeneration.from_pretrained(
     trust_remote_code=True,
     torch_dtype=torch.float16
 ).to(device).eval()
-# --- FIX: Set pad_token_id to silence the warning ---
-if model_m.config.pad_token_id is None:
-    model_m.config.pad_token_id = model_m.config.eos_token_id
-
 
 # Load Nanonets-OCR2-1.5B-exp
 MODEL_ID_N = "strangervisionhf/excess_layer_pruned-nanonets-1.5b"
@@ -152,9 +148,6 @@ model_n = AutoModelForImageTextToText.from_pretrained(
     torch_dtype=torch.float16,
     attn_implementation="flash_attention_2"
 ).to(device).eval()
-# --- FIX: Set pad_token_id to silence the warning ---
-if model_n.config.pad_token_id is None:
-    model_n.config.pad_token_id = model_n.config.eos_token_id
 
 
 # Load Dots.OCR from the local, patched directory
@@ -167,10 +160,6 @@ model_d = AutoModelForCausalLM.from_pretrained(
     device_map="auto",
     trust_remote_code=True
 ).eval()
-# --- FIX: Set pad_token_id to silence the warning ---
-if model_d.config.pad_token_id is None:
-    model_d.config.pad_token_id = model_d.config.eos_token_id
-
 
 # Load PaddleOCR
 MODEL_ID_P = "strangervisionhf/paddle"
@@ -180,9 +169,6 @@ model_p = AutoModelForCausalLM.from_pretrained(
     trust_remote_code=True,
     torch_dtype=torch.bfloat16
 ).to(device).eval()
-# --- FIX: Set pad_token_id to silence the warning ---
-if model_p.config.pad_token_id is None:
-    model_p.config.pad_token_id = model_p.config.eos_token_id
 
 @spaces.GPU
 def generate_image(model_name: str, text: str, image: Image.Image,
@@ -286,4 +272,4 @@ with gr.Blocks(css=css, theme=steel_blue_theme) as demo:
     )
 
 if __name__ == "__main__":
-    demo.queue(max_size=50).launch(show_error=True)
+    demo.queue(max_size=50).launch(mcp_server=True, ssr_mode=False, show_error=True)