Update app.py

app.py

@@ -1,90 +1,109 @@
 import gradio as gr
-from transformers import AutoProcessor, AutoModelForImageTextToText
+from transformers import AutoProcessor, AutoModel
 import torch
 from PIL import Image
 
-# 1. Konfigurasi
+# --- KONFIGURASI ---
 MODEL_PATH = "zai-org/GLM-OCR"
 
-# Paksa deteksi device yang aman
+# 1. Deteksi Hardware (Biar gak maksain CPU nangis)
 if torch.cuda.is_available():
     device = "cuda"
     dtype = torch.float16
 else:
     device = "cpu"
+    # Pake float32 aja buat CPU biar aman, walau lambat
     dtype = torch.float32
 
-print(f"⚡ Loading model ke: {device} dengan tipe data: {dtype}")
+print(f"🚀 Mulai System: Device={device} | Dtype={dtype}")
 
-# 2. Load Model & Processor
-# Kita load model dulu baru processor biar manajemen memori lebih rapi di Space gratisan
+# 2. LOAD MODEL (LOGIKA BARU)
+# GLM-OCR butuh arsitektur khusus. Kita pake 'AutoModel' bukan 'AutoModelForImageTextToText' 
+# karena 'AutoModel' lebih fleksibel buat nerima arsitektur custom via remote code.
 try:
-    model = AutoModelForImageTextToText.from_pretrained(
+    print("⏳ Loading Processor...")
+    processor = AutoProcessor.from_pretrained(
+        MODEL_PATH, 
+        trust_remote_code=True
+    )
+
+    print("⏳ Loading Model...")
+    # Pake AutoModel biasa, dia bakal baca config.json dan narik class GLMOCRModel otomatis
+    model = AutoModel.from_pretrained(
         MODEL_PATH,
         torch_dtype=dtype,
-        device_map="auto",
         trust_remote_code=True,
-        low_cpu_mem_usage=True # IQ move buat hemat RAM
+        # Opsi memori buat CPU (PENTING BUAT SPACE GRATISAN)
+        device_map="auto" if device == "cuda" else "cpu", 
+        low_cpu_mem_usage=True 
     )
     
-    # Kalo device cpu, pastikan model di float32
-    if device == "cpu":
-        model = model.float()
+    # Kalau CPU, model di eval mode
+    model = model.eval()
 
-    processor = AutoProcessor.from_pretrained(
-        MODEL_PATH, 
-        trust_remote_code=True
-    )
 except Exception as e:
-    print(f"❌ FATAL ERROR LOADING MODEL: {e}")
-    raise e
+    print(f"❌ KEGAGALAN SISTEM: {e}")
+    # Pesan ini biar muncul di log kalau error
+    raise ValueError(f"Gagal Load Model GLM-OCR. Detail: {e}")
 
-# 3. Logic Inferensi
-def run_ocr(image):
+# 3. Fungsi Kerja (Inferensi)
+def proses_gambar(image):
     if image is None:
-        return "⚠️ Tolong upload gambar dokumennya dulu, Bang."
+        return "⚠️ Woi bro, upload gambarnya dulu dong."
 
-    # Format Prompt
-    messages = [
+    # Prompt standar GLM-OCR
+    pesan_user = [
         {
             "role": "user",
             "content": [
                 {"type": "image", "image": image},
                 {"type": "text", "text": "Text Recognition:"}
-            ],
+            ]
         }
     ]
 
-    # Proses
     try:
+        # Preprocessing input
         inputs = processor.apply_chat_template(
-            messages,
+            pesan_user,
             add_generation_prompt=True,
             return_dict=True,
             return_tensors="pt"
         ).to(model.device)
 
-        # Generate (Batasi token biar gak timeout di CPU)
+        # Generasi Teks
         with torch.no_grad():
-            generated_ids = model.generate(**inputs, max_new_tokens=1024, do_sample=False)
-            
-        output_text = processor.decode(generated_ids[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True)
-        return output_text
+            output_ids = model.generate(
+                **inputs, 
+                max_new_tokens=1024, # Biar ga kepanjangan render di CPU
+                do_sample=False      # Greedy search biar stabil
+            )
+
+        # Ambil hasilnya doang (potong input dari output)
+        hasil_asli = output_ids[0][len(inputs["input_ids"][0]):]
+        teks_final = processor.decode(hasil_asli, skip_special_tokens=True)
+        
+        return teks_final
 
     except Exception as e:
-        return f"Error saat proses OCR: {str(e)}"
+        return f"😭 Error Pas Baca Gambar: {str(e)}"
 
-# 4. Interface (Gradio)
-with gr.Blocks(title="GLM-OCR by IQ 10M") as demo:
-    gr.Markdown("# 👁️ GLM-OCR Demo (CPU/GPU Friendly)")
-    gr.Markdown("Upload gambar dokumen. Sabar ya kalau pake CPU (Gratis), prosesnya agak lama.")
+# 4. Tampilan Web (UI)
+with gr.Blocks(theme=gr.themes.Soft()) as app:
+    gr.Markdown("# 🔍 GLM-OCR Detector (Versi IQ Tinggi)")
+    gr.Markdown("Model OCR canggih buat baca dokumen rumit. Upload aja langsung scan.")
     
     with gr.Row():
-        input_img = gr.Image(type="pil", label="Upload Disini", sources=["upload", "clipboard"])
-        output_txt = gr.TextArea(label="Hasil OCR", interactive=False)
+        with gr.Column():
+            img_input = gr.Image(type="pil", label="Gambar Dokumen/Teks", sources=["upload", "clipboard"])
+            scan_btn = gr.Button("🚀 MULAI SCAN", variant="primary")
+        
+        with gr.Column():
+            txt_output = gr.TextArea(label="Hasil Bacaan", show_copy_button=True, lines=20)
             
-    btn = gr.Button("�� Scan Dokumen", variant="primary")
-    btn.click(fn=run_ocr, inputs=input_img, outputs=output_txt)
+    scan_btn.click(fn=proses_gambar, inputs=img_input, outputs=txt_output)
 
+# 5. Eksekusi
 if __name__ == "__main__":
-    demo.launch()
+    print("✅ App siap dijalankan...")
+    app.launch()