Update backend/api.py

backend/api.py

@@ -1,45 +1,42 @@
 from fastapi import FastAPI
 from fastapi.middleware.cors import CORSMiddleware
 from pydantic import BaseModel
-from transformers import AutoTokenizer, AutoModelForSequenceClassification
+from transformers import AutoTokenizer
+import onnxruntime as ort
 import torch
 import pandas as pd
 from pathlib import Path
 
 app = FastAPI()
 
-# === CORS agar bisa diakses frontend Vercel ===
+# === CORS untuk frontend di Vercel ===
 app.add_middleware(
     CORSMiddleware,
-    allow_origins=["*"],  # bisa diganti domain spesifik
+    allow_origins=["*"],  
     allow_credentials=True,
     allow_methods=["*"],
     allow_headers=["*"],
 )
 
-# === Global Setup ===
+# === Path setup ===
 BASE_DIR = Path(__file__).resolve().parent
-MODEL_DIR = BASE_DIR / "bert_chatbot_model"
+MODEL_PATH = BASE_DIR / "bert_chatbot_model.onnx"
+TOKENIZER_PATH = BASE_DIR / "bert_chatbot_tokenizer"
 DATASET_PATH = BASE_DIR / "dataset_chatbot_template.xlsx"
 
-print("🚀 Loading model...")
-try:
-    tokenizer = AutoTokenizer.from_pretrained(str(MODEL_DIR))
-    model = AutoModelForSequenceClassification.from_pretrained(str(MODEL_DIR))
-    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-    model.to(device)
-    model.eval()
-    print("✅ Model loaded successfully!")
-except Exception as e:
-    print(f"❌ Model load error: {e}")
+# === 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!")
 
-# === Load Dataset (optional untuk custom response) ===
+# === Load dataset (optional) ===
 try:
     df_jawaban = pd.read_excel(DATASET_PATH)
 except Exception:
     df_jawaban = pd.DataFrame(columns=["Intent", "Jawaban_ID"])
 
-# === Default Responses ===
+# === 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.",
@@ -50,31 +47,45 @@ responses = {
     "fallback": "I'm sorry, I don't understand. Please try another question."
 }
 
+# === Request schema ===
 class ChatRequest(BaseModel):
     text: str
 
 @app.get("/")
 async def root():
-    return {"message": "🚀 Chatbot API running on Hugging Face"}
+    return {"message": "🚀 ONNX Chatbot API running on Hugging Face"}
 
 @app.post("/chatbot")
-async def chat(req: ChatRequest):
+async def chatbot(req: ChatRequest):
     try:
-        inputs = tokenizer(req.text, return_tensors="pt", padding=True, truncation=True, max_length=128).to(device)
-        with torch.no_grad():
-            outputs = model(**inputs)
-            pred_id = torch.argmax(outputs.logits, dim=1).item()
+        # 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()}
+        ort_outputs = session.run(None, ort_inputs)
+        logits = torch.tensor(ort_outputs[0])
+        pred_id = torch.argmax(logits, dim=1).item()
 
-        intent = model.config.id2label.get(pred_id, "fallback")
+        # === Mapping ID ke label ===
+        id2label = {
+            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:
             reply = df_jawaban.loc[df_jawaban["Intent"] == intent, "Jawaban_ID"].iloc[0]
         else:
             reply = responses.get(intent, responses["fallback"])
 
-        # ✅ agar cocok dengan frontend
         return {"reply": reply, "intent": intent}
 
     except Exception as e:
-        print(f"❌ Runtime Error: {e}")
+        print(f"❌ Runtime error: {e}")
         return {"reply": "⚠️ Internal server error.", "intent": "error"}