Update inference_test_turkcell_with_intents.py

inference_test_turkcell_with_intents.py

@@ -75,40 +75,48 @@ def root():
     </body></html>
     """
 
-@app.post("/train_intents")
+@app.post("/train_intents", status_code=202)
 def train_intents(train_input: TrainInput):
     global INTENT_DEFINITIONS
-    try:
-        intents = train_input.intents
-        INTENT_DEFINITIONS = {intent["name"]: intent for intent in intents}
-        texts, labels, label2id = [], [], {}
-        for idx, intent in enumerate(intents):
-            label2id[intent["name"]] = idx
-            for ex in intent["examples"]:
-                texts.append(ex)
-                labels.append(idx)
-
-        dataset = Dataset.from_dict({"text": texts, "label": labels})
-        tokenizer = AutoTokenizer.from_pretrained(INTENT_MODEL_ID)
-        model = AutoModelForSequenceClassification.from_pretrained(INTENT_MODEL_ID, num_labels=len(label2id))
-
-        def tokenize(batch):
-            return tokenizer(batch["text"], truncation=True, padding=True)
-
-        tokenized = dataset.map(tokenize, batched=True)
-        args = TrainingArguments("./intent_train_output", per_device_train_batch_size=4, num_train_epochs=3, logging_steps=10, save_strategy="no", report_to=[])
-        trainer = Trainer(model=model, args=args, train_dataset=tokenized)
-        trainer.train()
-
-        if os.path.exists(INTENT_MODEL_PATH): shutil.rmtree(INTENT_MODEL_PATH)
-        model.save_pretrained(INTENT_MODEL_PATH)
-        tokenizer.save_pretrained(INTENT_MODEL_PATH)
-        with open(os.path.join(INTENT_MODEL_PATH, "label2id.json"), "w") as f:
-            json.dump(label2id, f)
-
-        return {"status": "ok", "message": "Intent modeli eğitildi."}
-    except Exception as e:
-        return JSONResponse(content={"error": str(e)}, status_code=500)
+    log("📥 POST /train_intents çağrıldı.")
+    intents = train_input.intents
+    INTENT_DEFINITIONS = {intent["name"]: intent for intent in intents}
+
+    def background_training():
+        try:
+            log("🔧 Intent eğitimi başlatıldı...")
+            texts, labels, label2id = [], [], {}
+            for idx, intent in enumerate(intents):
+                label2id[intent["name"]] = idx
+                for ex in intent["examples"]:
+                    texts.append(ex)
+                    labels.append(idx)
+
+            dataset = Dataset.from_dict({"text": texts, "label": labels})
+            tokenizer = AutoTokenizer.from_pretrained(INTENT_MODEL_ID)
+            model = AutoModelForSequenceClassification.from_pretrained(INTENT_MODEL_ID, num_labels=len(label2id))
+
+            def tokenize(batch):
+                return tokenizer(batch["text"], truncation=True, padding=True)
+
+            tokenized = dataset.map(tokenize, batched=True)
+            args = TrainingArguments("./intent_train_output", per_device_train_batch_size=4, num_train_epochs=3, logging_steps=10, save_strategy="no", report_to=[])
+            trainer = Trainer(model=model, args=args, train_dataset=tokenized)
+            trainer.train()
+
+            if os.path.exists(INTENT_MODEL_PATH): shutil.rmtree(INTENT_MODEL_PATH)
+            model.save_pretrained(INTENT_MODEL_PATH)
+            tokenizer.save_pretrained(INTENT_MODEL_PATH)
+            with open(os.path.join(INTENT_MODEL_PATH, "label2id.json"), "w") as f:
+                json.dump(label2id, f)
+
+            log("✅ Intent eğitimi tamamlandı ve model kaydedildi.")
+        except Exception as e:
+            log(f"❌ Intent eğitimi hatası: {e}")
+            traceback.print_exc()
+
+    threading.Thread(target=background_training, daemon=True).start()
+    return {"status": "accepted", "message": "Intent eğitimi arka planda başlatıldı."}
 
 @app.post("/load_intent_model")
 def load_intent_model():