Update backend/api.py

backend/api.py

@@ -8,38 +8,42 @@ from pathlib import Path
 import traceback
 
 # === Inisialisasi FastAPI ===
-app = FastAPI()
+app = FastAPI(title="Portfolio Chatbot API", version="1.0")
 
-# === CORS untuk frontend (misal dari Vercel) ===
+# === CORS (boleh dibatasi ke domain Vercel kamu nanti) ===
 app.add_middleware(
     CORSMiddleware,
-    allow_origins=["*"],      # Bisa dibatasi ke domain Vercel kamu untuk keamanan
+    allow_origins=["*"],  # contoh: ["https://your-frontend.vercel.app"]
     allow_credentials=True,
     allow_methods=["*"],
     allow_headers=["*"],
 )
 
-# === Path ===
+# === Path model dan tokenizer ===
 BASE_DIR = Path(__file__).resolve().parent
 MODEL_PATH = BASE_DIR / "models" / "bert_chatbot.onnx"
 TOKENIZER_PATH = BASE_DIR / "models" / "bert-base-multilingual-cased"
 
-# === Global variable ===
+# === Global variable untuk model dan tokenizer ===
 tokenizer = None
 session = None
 
 # === Load model dan tokenizer ===
-try:
-    print("🚀 Loading tokenizer dan ONNX model...")
-    tokenizer = AutoTokenizer.from_pretrained(str(TOKENIZER_PATH))
-    session = ort.InferenceSession(str(MODEL_PATH), providers=["CPUExecutionProvider"])
-    print("✅ Model dan tokenizer berhasil dimuat!")
-    print("📥 Model expects inputs:", [i.name for i in session.get_inputs()])
-except Exception as e:
-    print("❌ ERROR saat memuat model/tokenizer:", e)
-    traceback.print_exc()
-
-# === Label mapping (HARUS SAMA DENGAN TRAINING) ===
+def load_model():
+    global tokenizer, session
+    try:
+        print("🚀 Loading tokenizer dan ONNX model...")
+        tokenizer = AutoTokenizer.from_pretrained(str(TOKENIZER_PATH))
+        session = ort.InferenceSession(str(MODEL_PATH), providers=["CPUExecutionProvider"])
+        print("✅ Model dan tokenizer berhasil dimuat!")
+        print("📥 Model expects inputs:", [i.name for i in session.get_inputs()])
+    except Exception as e:
+        print("❌ ERROR saat memuat model/tokenizer:", e)
+        traceback.print_exc()
+
+load_model()
+
+# === Label mapping (HARUS sama seperti saat training) ===
 id2label = {
     0: "about_me",
     1: "career_goal",
@@ -50,7 +54,7 @@ id2label = {
     6: "skills",
 }
 
-# === Kamus respon SAMA dengan training ===
+# === Kamus respon sesuai training ===
 responses = {
     "about_me": "I am a passionate developer specializing in AI and web development.",
     "skills": "My main skills are HTML5, CSS3, JavaScript, Laravel, Node.js, Database, TensorFlow, PyTorch, Firebase, and Jupyter Notebook.",
@@ -65,32 +69,46 @@ responses = {
 class ChatRequest(BaseModel):
     text: str
 
+# === Root endpoint ===
 @app.get("/")
 async def root():
-    return {"message": "🚀 Chatbot API (ONNX) running on Hugging Face"}
+    return {"message": "🚀 Portfolio Chatbot API is running successfully!"}
 
+# === Chatbot endpoint ===
 @app.post("/chatbot")
 async def chatbot(req: ChatRequest):
+    """
+    Endpoint utama untuk memproses input teks dan mengembalikan intent serta respon.
+    """
     intent = "fallback"
 
-    if session is None:
+    # Pastikan model sudah termuat
+    if session is None or tokenizer is None:
         return {"reply": responses["fallback"], "intent": "error_loading"}
 
     try:
         # === Tokenisasi input ===
-        inputs = tokenizer(req.text, return_tensors="np", padding=True, truncation=True, max_length=128)
-
-        # === Ambil hanya input yang cocok dengan model ONNX ===
+        inputs = tokenizer(
+            req.text,
+            return_tensors="np",  # output dalam format numpy
+            padding=True,
+            truncation=True,
+            max_length=128
+        )
+
+        # === Siapkan input sesuai nama yang diminta oleh model ===
         expected_inputs = [i.name for i in session.get_inputs()]
         ort_inputs = {k: v.astype(np.int64) for k, v in inputs.items() if k in expected_inputs}
 
-        # === Inferensi ONNX ===
+        # === Jalankan inferensi ONNX ===
         ort_outputs = session.run(None, ort_inputs)
         logits = ort_outputs[0]
-        pred_id = np.argmax(logits, axis=1)[0]
 
-        # === Ambil intent & respon ===
+        # === Prediksi intent ===
+        pred_id = int(np.argmax(logits, axis=1)[0])
         intent = id2label.get(pred_id, "fallback")
+
+        # === Ambil respon ===
         reply = responses.get(intent, responses["fallback"])
 
         print(f"🧠 Input: {req.text} | Intent: {intent} | Reply: {reply}")