Update backend/api.py

backend/api.py

@@ -3,16 +3,17 @@ from fastapi.middleware.cors import CORSMiddleware
 from pydantic import BaseModel
 from transformers import AutoTokenizer
 import onnxruntime as ort
-import torch
+import numpy as np # Menggantikan torch untuk operasi array
 import pandas as pd
 from pathlib import Path
+import traceback # Untuk debugging di log
 
 app = FastAPI()
 
 # === CORS untuk frontend di Vercel ===
 app.add_middleware(
     CORSMiddleware,
-    allow_origins=["*"],  
+    allow_origins=["*"],      
     allow_credentials=True,
     allow_methods=["*"],
     allow_headers=["*"],
@@ -20,27 +21,47 @@ app.add_middleware(
 
 # === Path setup ===
 BASE_DIR = Path(__file__).resolve().parent
+
+# GANTI DENGAN NAMA FILE ONNX YANG SESUAI DI FOLDER 'models'
+# Jika file Anda bernama bert_chatbot.onnx dan berada di models/
 MODEL_PATH = BASE_DIR / "models" / "bert_chatbot.onnx"
-TOKENIZER_PATH = BASE_DIR / "models" / "bert-base-multilingual-cased"
+TOKENIZER_PATH = BASE_DIR / "models" / "bert-base-multilingual-cased" 
 DATASET_PATH = BASE_DIR / "dataset_chatbot_template.xlsx"
 
-# === Load tokenizer dan model ===
-print("🚀 Loading ONNX model...")
-tokenizer = AutoTokenizer.from_pretrained(str(TOKENIZER_PATH))
-session = ort.InferenceSession(str(MODEL_PATH), providers=["CPUExecutionProvider"])
-print("✅ ONNX model loaded!")
+# === Global Variables ===
+tokenizer = None
+session = None
+df_jawaban = None
 
-# === Load dataset (optional) ===
+# === Load tokenizer dan model ===
 try:
+    print("🚀 Loading ONNX model...")
+    
+    # 1. Muat Tokenizer
+    tokenizer = AutoTokenizer.from_pretrained(str(TOKENIZER_PATH))
+    
+    # 2. Muat ONNX Runtime Session (Provider CPU adalah yang paling stabil di HF Free Tier)
+    session = ort.InferenceSession(str(MODEL_PATH), providers=["CPUExecutionProvider"])
+    
+    # 3. Muat Dataset
     df_jawaban = pd.read_excel(DATASET_PATH)
-except Exception:
-    df_jawaban = pd.DataFrame(columns=["Intent", "Jawaban_ID"])
+    
+    print("✅ ONNX model loaded!")
+
+except Exception as e:
+    # Ini akan mencetak error jika path salah atau file tidak ditemukan
+    print("--------------------------------------------------")
+    print(f"❌ FATAL ERROR SAAT MEMUAT ONNX/SUMBER DAYA: {e}")
+    traceback.print_exc() 
+    print("--------------------------------------------------")
+    pass
+
 
 # === Default responses ===
 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, TensorFlow, and PyTorch.",
-    "projects": "Some of my projects include Bald Detection and Portfolio Website.",
+    "projects": "Some of my projects include Mobile Apps Bald Detection and Portfolio Website.",
     "experience": "I have worked as IT Support, AI Engineer, and Freelancer.",
     "career_goal": "My career goal is to become a Full Stack Developer and ML Engineer.",
     "greeting": "Hello! How can I help you regarding this portfolio?",
@@ -57,35 +78,46 @@ async def root():
 
 @app.post("/chatbot")
 async def chatbot(req: ChatRequest):
+    # Cek jika session ONNX gagal dimuat saat startup
+    if session is None:
+        return {"reply": responses["fallback"], "intent": "error_loading"}
+        
     try:
-        # Tokenize input
-        inputs = tokenizer(req.text, return_tensors="pt", padding=True, truncation=True, max_length=128)
-
-        # Convert to numpy for ONNX
-        ort_inputs = {k: v.cpu().numpy() for k, v in inputs.items()}
+        # 1. Tokenisasi (return_tensors="np" karena kita menggunakan NumPy/ONNX)
+        # return_tensors="np" menghemat konversi PyTorch
+        inputs = tokenizer(req.text, return_tensors="np", padding=True, truncation=True, max_length=128)
+        
+        # 2. Dapatkan nama input dari ONNX Session
+        input_names = [i.name for i in session.get_inputs()]
+        
+        # Mapping input NumPy ke nama input ONNX
+        ort_inputs = {name: inputs[name] for name in input_names}
+        
+        # 3. Inferensi ONNX
         ort_outputs = session.run(None, ort_inputs)
-        logits = torch.tensor(ort_outputs[0])
-        pred_id = torch.argmax(logits, dim=1).item()
-
-        # === Mapping ID ke label ===
+        
+        # 4. Ambil Logit dan Prediksi (numpy)
+        logits = ort_outputs[0]
+        pred_id = np.argmax(logits, axis=1)[0]
+        
+        # === Mapping ID ke label (ANDA HARUS TAHU MAPPING INI) ===
         id2label = {
-            0: "about_me",
-            1: "skills",
-            2: "projects",
-            3: "experience",
-            4: "career_goal",
-            5: "greeting",
+            0: "about_me", 1: "skills", 2: "projects", 3: "experience", 
+            4: "career_goal", 5: "greeting",
         }
         intent = id2label.get(pred_id, "fallback")
-
+        
         # === Ambil jawaban ===
         if not df_jawaban.empty and intent in df_jawaban["Intent"].values:
+            # Menggunakan .astype(str) untuk mencegah error Pandas tipe data
             reply = df_jawaban.loc[df_jawaban["Intent"] == intent, "Jawaban_ID"].iloc[0]
         else:
             reply = responses.get(intent, responses["fallback"])
-
-        return {"reply": reply, "intent": intent}
-
+            
+        return {"reply": str(reply), "intent": intent}
+        
     except Exception as e:
+        import traceback
         print(f"❌ Runtime error: {e}")
-        return {"reply": "⚠️ Internal server error.", "intent": "error"}
+        traceback.print_exc()
+        return {"reply": "⚠️ Internal server error.", "intent": "error"}