Robust Fix: Pin env to Python 3.9 and Gradio 4.30

README.md

@@ -16,8 +16,8 @@ tags:
 datasets:
 - c3rl/IIIT-INDIC-HW-WORDS-Hindi
 sdk: gradio
-sdk_version: 4.44.1
-python_version: "3.10"
+sdk_version: 4.30.0
+python_version: 3.9
 app_file: app.py
 pinned: true
 ---

app.py

@@ -6,14 +6,15 @@ from transformers import AutoTokenizer, TrOCRProcessor, ViTImageProcessor, Visio
 
 # Configuration
 BASE_MODEL_ID = "paudelanil/trocr-devanagari-2"
+# Ensuring the HF Username is correct
 ADAPTER_ID = "manishw10/devgen-trocr-devanagari-lora" 
 
-print("Loading model...")
+print("System: Loading model components...")
 device = "cuda" if torch.cuda.is_available() else "cpu"
 
 try:
     processor = TrOCRProcessor.from_pretrained(BASE_MODEL_ID)
-except:
+except Exception:
     image_processor = ViTImageProcessor.from_pretrained("google/vit-base-patch16-224-in21k")
     tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL_ID)
     processor = TrOCRProcessor(image_processor=image_processor, tokenizer=tokenizer)
@@ -22,28 +23,33 @@ base_model = VisionEncoderDecoderModel.from_pretrained(BASE_MODEL_ID)
 model = PeftModel.from_pretrained(base_model, ADAPTER_ID)
 model.to(device)
 model.eval()
-print(f"Model loaded on {device}")
+print(f"System: Model loaded successfully on {device}")
 
 def predict(image):
     if image is None:
-        return "Please upload an image."
-    image = image.convert("RGB")
-    pixel_values = processor(image, return_tensors="pt").pixel_values.to(device)
-    with torch.no_grad():
-        generated_ids = model.generate(pixel_values)
-        generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
-    return generated_text
+        return "Error: No image uploaded"
+    try:
+        image = image.convert("RGB")
+        pixel_values = processor(image, return_tensors="pt").pixel_values.to(device)
+        with torch.no_grad():
+            generated_ids = model.generate(pixel_values)
+            generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
+        return generated_text
+    except Exception as e:
+        return f"Error during inference: {str(e)}"
 
-# Simplified Interface without Examples to avoid the Gradio 4.x schema bug
+# The most stable UI structure: gr.Interface
+# This avoids the complex nested blocks that trigger the Gradio 4.x schema bug
 demo = gr.Interface(
     fn=predict,
-    inputs=gr.Image(type="pil", label="Upload Handwritten Devanagari Word"),
+    inputs=gr.Image(type="pil", label="Devanagari Image"),
     outputs=gr.Textbox(label="Recognized Text"),
-    title="DevGen Devanagari OCR",
-    description="High-precision handwritten word recognition powered by TrOCR & LoRA. Upload a cropped image of a Devanagari word to recognize the text.",
+    title="DevGen OCR Demo",
+    description="Handwritten Devanagari word recognition.",
     allow_flagging="never"
 )
 
 if __name__ == "__main__":
-    # Standard HF Spaces launch
-    demo.launch(server_name="0.0.0.0", server_port=7860)
+    # Robust launch: No manual IP/Port. 
+    # Gradio automatically configures itself for Hugging Face Spaces.
+    demo.launch()

requirements.txt

@@ -5,3 +5,5 @@ huggingface_hub==0.25.2
 pillow
 safetensors
 sentencepiece
+fastapi<0.113.0
+uvicorn