Upload app.py with huggingface_hub

app.py

@@ -1,5 +1,9 @@
 
 import os
+import warnings
+# Hide annoying PyTorch deprecation warnings
+warnings.filterwarnings("ignore")
+
 import gradio as gr
 import torch
 import torch.nn as nn
@@ -8,7 +12,7 @@ from huggingface_hub import hf_hub_download
 import json
 import string
 
-MAX_SEQ_LEN = 2000 # Updated for Phase 3 large JSON 
+MAX_SEQ_LEN = 2000
 
 class CSMTokenizer:
     def __init__(self):
@@ -21,7 +25,6 @@ class CSMTokenizer:
     def decode(self, tokens):
         return "".join([self.inverse_vocab.get(t, "") for t in tokens if t not in [self.PAD, self.SOS, self.EOS]])
 
-# UPGRADED ARCHITECTURE MATCHING PHASE 3
 class CSMVisionEncoder(nn.Module):
     def __init__(self, embed_dim=512):
         super().__init__()
@@ -57,7 +60,7 @@ class CSMNativeModel(nn.Module):
         self.decoder = CSMTextDecoder(vocab_size)
 
 tokenizer = CSMTokenizer()
-device = torch.device("cpu") # Force CPU for free spaces
+device = torch.device("cpu")
 
 print("Downloading Final Production Model Phase 3...")
 HF_SECURE_TOKEN = os.environ.get("HF_TOKEN")
@@ -65,7 +68,6 @@ HF_SECURE_TOKEN = os.environ.get("HF_TOKEN")
 model_path = hf_hub_download(repo_id="Chhagan005/CSM-KIE-Universal", filename="csm_kie_model.pth", token=HF_SECURE_TOKEN)
 model = CSMNativeModel(tokenizer.vocab_size)
 
-# Apply quantization structure to match the loaded weights
 import torch.ao.quantization 
 model = torch.ao.quantization.quantize_dynamic(model, {nn.Linear, nn.Conv2d}, dtype=torch.qint8)
 model.load_state_dict(torch.load(model_path, map_location=device))
@@ -104,7 +106,7 @@ def process_id_card(front_img, back_img):
 
 with gr.Blocks() as demo:
     gr.Markdown("# 🪪 CSM-KIE Master VLM Scanner")
-    gr.Markdown("Production Mode: Phase 3 Foundation Architecture. Extracts fully structured dynamic JSON data from International ID cards (Middle East, Africa, etc.)")
+    gr.Markdown("Production Mode: Phase 3 Foundation Architecture. Extracts fully structured dynamic JSON data from International ID cards.")
     
     with gr.Row():
         with gr.Column():
@@ -117,4 +119,6 @@ with gr.Blocks() as demo:
             
     scan_btn.click(process_id_card, inputs=[front, back], outputs=output_json)
 
-demo.launch(theme="soft")
+# FIX: Forcing Port Binding for Hugging Face Spaces
+if __name__ == "__main__":
+    demo.launch(server_name="0.0.0.0", server_port=7860)