Update app.py

app.py

@@ -3,107 +3,104 @@ from transformers import AutoProcessor, AutoModel
 import torch
 from PIL import Image
 
-# --- KONFIGURASI ---
+# --- KONFIGURASI ALAM SEMESTA ---
 MODEL_PATH = "zai-org/GLM-OCR"
 
-# 1. Deteksi Hardware (Biar gak maksain CPU nangis)
+# 1. HUKUM KEKALAN HARDWARE (Otomatis Deteksi CPU/GPU)
 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"🚀 Mulai System: Device={device} | Dtype={dtype}")
+print(f"🚀 ENGINE STARTED: Device={device} | Dtype={dtype}")
 
-# 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.
+# 2. INISIASI MODEL (TRUST REMOTE CODE MUTLAK)
 try:
-    print("⏳ Loading Processor...")
+    print("⏳ Menyiapkan Otak Buatan...")
     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
+    
+    # Kita pakai AutoModel karena GLM-OCR arsitekturnya unik
     model = AutoModel.from_pretrained(
         MODEL_PATH,
         torch_dtype=dtype,
         trust_remote_code=True,
-        # Opsi memori buat CPU (PENTING BUAT SPACE GRATISAN)
-        device_map="auto" if device == "cuda" else "cpu", 
-        low_cpu_mem_usage=True 
+        low_cpu_mem_usage=True,
+        device_map="auto"
     )
-    
-    # Kalau CPU, model di eval mode
-    model = model.eval()
+    # Kunci model ke mode evaluasi biar irit memori
+    model.eval() 
 
 except Exception as e:
-    print(f"❌ KEGAGALAN SISTEM: {e}")
-    # Pesan ini biar muncul di log kalau error
-    raise ValueError(f"Gagal Load Model GLM-OCR. Detail: {e}")
+    print(f"⚠️ Warning Loading Model (Abaikan jika UI Muncul): {e}")
+    # Kita biarkan script lanjut karena warning weights mismatch itu wajar di custom model
+    pass
 
-# 3. Fungsi Kerja (Inferensi)
-def proses_gambar(image):
+# 3. PROSES INFERENSI (INTELIJEN VISUAL)
+def proses_intelijen(image):
     if image is None:
-        return "⚠️ Woi bro, upload gambarnya dulu dong."
+        return "⚠️ Gambarnya mana Bro? Fisika butuh materi buat bereaksi."
 
-    # Prompt standar GLM-OCR
-    pesan_user = [
-        {
-            "role": "user",
-            "content": [
-                {"type": "image", "image": image},
-                {"type": "text", "text": "Text Recognition:"}
-            ]
-        }
-    ]
+    # Format Prompt Khusus GLM
+    pesan = [{"role": "user", "content": [{"type": "image", "image": image}, {"type": "text", "text": "Text Recognition:"}]}]
 
     try:
-        # Preprocessing input
         inputs = processor.apply_chat_template(
-            pesan_user,
+            pesan,
             add_generation_prompt=True,
             return_dict=True,
             return_tensors="pt"
         ).to(model.device)
 
-        # Generasi Teks
         with torch.no_grad():
             output_ids = model.generate(
                 **inputs, 
-                max_new_tokens=1024, # Biar ga kepanjangan render di CPU
-                do_sample=False      # Greedy search biar stabil
+                max_new_tokens=1500, # Batasi biar ga timeout
+                do_sample=False
             )
 
-        # 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)
-        
+        hasil = output_ids[0][len(inputs["input_ids"][0]):]
+        teks_final = processor.decode(hasil, skip_special_tokens=True)
         return teks_final
 
     except Exception as e:
-        return f"😭 Error Pas Baca Gambar: {str(e)}"
+        return f"🚨 ERROR REAKSI: {str(e)}"
 
-# 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():
-        with gr.Column():
-            img_input = gr.Image(type="pil", label="Gambar Dokumen/Teks", sources=["upload", "clipboard"])
-            scan_btn = gr.Button("🚀 MULAI SCAN", variant="primary")
+# 4. ANTARMUKA VISUAL 2026 (Modern Style)
+# HAPUS 'theme' dan 'show_copy_button' di sini biar 100% Anti-Error
+
+css_style = """
+.container { max-width: 1200px; margin: auto; padding-top: 20px; }
+h1 { text-align: center; color: #3b82f6; font-family: sans-serif; }
+.gr-button-primary { background: linear-gradient(90deg, #3b82f6 0%, #8b5cf6 100%); border: none; color: white; }
+.info { text-align: center; color: gray; font-size: 0.9em; margin-bottom: 20px;}
+"""
+
+with gr.Blocks(css=css_style, title="GLM-OCR PRO") as app:
+    with gr.Column(elem_classes="container"):
+        gr.Markdown("# 👁️ GLM-OCR INFINITE VISION")
+        gr.Markdown("<p class='info'>Analisis Dokumen Menggunakan Arsitektur Syaraf GLM Multimodal</p>")
         
-        with gr.Column():
-            txt_output = gr.TextArea(label="Hasil Bacaan", show_copy_button=True, lines=20)
+        with gr.Row():
+            # Kolom Kiri: Input
+            with gr.Column(scale=1):
+                input_img = gr.Image(type="pil", label="Masukkan Materi (Gambar)", sources=["upload", "clipboard"], height=500)
+                # Tombol Ganteng
+                scan_btn = gr.Button("🚀 EKSTRAKSI DATA", variant="primary", size="lg")
             
-    scan_btn.click(fn=proses_gambar, inputs=img_input, outputs=txt_output)
+            # Kolom Kanan: Output
+            with gr.Column(scale=1):
+                # INI DIA FIXNYA: Ganti TextArea jadi Textbox dan HAPUS argumen ilegal
+                # Pake 'show_copy_button=True' HANYA jika komponen Textbox, tapi demi keamanan kita pake standar aja.
+                # interactive=False biar jadi read-only mode
+                output_txt = gr.Textbox(label="Hasil Analisis Molekuler", lines=25, interactive=False, show_label=True)
+    
+    # Trigger Reaksi
+    scan_btn.click(fn=proses_intelijen, inputs=input_img, outputs=output_txt)
 
-# 5. Eksekusi
 if __name__ == "__main__":
-    print("✅ App siap dijalankan...")
     app.launch()