Spaces:

UcsTurkey
/

flare

Running

App Files Files Community

ciyidogan commited on May 23

Commit

9660152

verified ·

1 Parent(s): 8be9d0b

Update inference_test.py

Browse files

Files changed (1) hide show

inference_test.py +6 -23

inference_test.py CHANGED Viewed

@@ -4,7 +4,7 @@ from fastapi.responses import HTMLResponse, JSONResponse
 from transformers import AutoTokenizer, AutoModelForSequenceClassification
 import intent_test_runner
 from service_config import ServiceConfig
-import intent, intent, llm_model
 from log import log
 s_config = ServiceConfig()
@@ -12,14 +12,11 @@ s_config.setup_environment()
 # === FastAPI
 app = FastAPI()
-chat_history = []
 @app.get("/")
 def health():
     return {"status": "ok"}
-import uuid  # yukarıda zaten eklendiğini varsayıyoruz
 @app.post("/run_tests", status_code=202)
 def run_tests():
     log("🚦 /run_tests çağrıldı. Testler başlatılıyor...")
@@ -28,7 +25,6 @@ def run_tests():
 @app.get("/start", response_class=HTMLResponse)
 def root():
-    # Yeni session ID üret
     session_id = str(uuid.uuid4())
     session_info = {
         "session_id": session_id,
@@ -37,15 +33,10 @@ def root():
         "last_intent": None,
         "awaiting_variable": None
     }
-    # Session store başlatıldıysa ekle
     if not hasattr(app.state, "session_store"):
         app.state.session_store = {}
     app.state.session_store[session_id] = session_info
     log(f"🌐 /start ile yeni session başlatıldı: {session_id}")
-    # HTML + session_id gömülü
     return f"""
     <html><body>
         <h2>Turkcell LLM Chat</h2>
@@ -56,7 +47,6 @@ def root():
         <script>
         const sessionId = "{session_id}";
         localStorage.setItem("session_id", sessionId);
         async function send() {{
             const input = document.getElementById("input").value;
             const res = await fetch('/chat', {{
@@ -78,10 +68,8 @@ def root():
 def start_chat():
     if llm_model.model is None or llm_model.tokenizer is None:
         return {"error": "Model yüklenmedi."}
     if not hasattr(app.state, "session_store"):
         app.state.session_store = {}
     session_id = str(uuid.uuid4())
     session_info = {
         "session_id": session_id,
@@ -134,7 +122,6 @@ async def chat(msg: llm_model.Message, request: Request):
         if llm_model.model is None or llm_model.tokenizer is None:
             return {"error": "Model yüklenmedi."}
-        # 🎯 Intent modeli varsa her mesajda intent tespiti yap
         detected_intent, intent_conf = None, 0.0
         if s_config.INTENT_MODEL:
             detected_intent, intent_conf = await intent.detect_intent(user_input)
@@ -143,7 +130,6 @@ async def chat(msg: llm_model.Message, request: Request):
         current_intent = session.get("last_intent")
         awaiting_variable = session.get("awaiting_variable")
-        # 🧹 Kullanıcı farklı intent başlattıysa → context sıfırlanır
         if (
             awaiting_variable and
             detected_intent and
@@ -156,7 +142,6 @@ async def chat(msg: llm_model.Message, request: Request):
             session["last_intent"] = detected_intent
             current_intent = detected_intent
-        # 🎯 Eğer intent geçerliyse ve tanımlıysa, intent’e göre işleyişe gir
         if (
             detected_intent and
             intent_conf > s_config.INTENT_CONFIDENCE_THRESHOLD and
@@ -167,7 +152,6 @@ async def chat(msg: llm_model.Message, request: Request):
             data_formats = s_config.DATA_FORMATS
             variable_format_map = definition.get("variable_formats", {})
-            # 🧩 Awaiting variable varsa onu çözmeye çalış
             if awaiting_variable:
                 extracted = intent.extract_parameters(pattern_list, user_input)
                 for p in extracted:
@@ -177,12 +161,10 @@ async def chat(msg: llm_model.Message, request: Request):
                         log(f"✅ Awaiting parametre tamamlandı: {awaiting_variable} = {p['value']}")
                         break
-            # 🔍 Tüm parametreleri yeniden değerlendir
             extracted = intent.extract_parameters(pattern_list, user_input)
             variables = {p["key"]: p["value"] for p in extracted}
             session.setdefault("variables", {}).update(variables)
-            # ✔️ Validasyon
             is_valid, validation_errors = intent.validate_variable_formats(session["variables"], variable_format_map, data_formats)
             if not is_valid:
                 session["awaiting_variable"] = list(validation_errors.keys())[0]
@@ -190,7 +172,6 @@ async def chat(msg: llm_model.Message, request: Request):
                 app.state.session_store = session_store
                 return {"response": list(validation_errors.values())[0]}
-            # ❓ Eksik parametre kontrolü
             expected_vars = list(variable_format_map.keys())
             missing_vars = [v for v in expected_vars if v not in session["variables"]]
             if missing_vars:
@@ -199,7 +180,6 @@ async def chat(msg: llm_model.Message, request: Request):
                 app.state.session_store = session_store
                 return {"response": f"Lütfen {missing_vars[0]} bilgisini belirtir misiniz?"}
-            # 🚀 Intent çalıştır
             result = intent.execute_intent(
                 detected_intent,
                 user_input,
@@ -218,7 +198,6 @@ async def chat(msg: llm_model.Message, request: Request):
             else:
                 return {"response": random.choice(s_config.FALLBACK_ANSWERS)}
-        # 🤖 Intent algılanamadıysa veya threshold altındaysa LLM’e sor
         session["awaiting_variable"] = None
         session["variables"] = {}
         response, response_conf = await llm_model.generate_response(user_input, s_config)
@@ -230,7 +209,11 @@ async def chat(msg: llm_model.Message, request: Request):
         traceback.print_exc()
         return JSONResponse(content={"error": str(e)}, status_code=500)
 threading.Thread(target=llm_model.setup_model, kwargs={"s_config": s_config}, daemon=True).start()
 threading.Thread(target=lambda: uvicorn.run(app, host="0.0.0.0", port=7860), daemon=True).start()
 while True:
-    time.sleep(60)

 from transformers import AutoTokenizer, AutoModelForSequenceClassification
 import intent_test_runner
 from service_config import ServiceConfig
+import intent, llm_model
 from log import log
 s_config = ServiceConfig()
 # === FastAPI
 app = FastAPI()
 @app.get("/")
 def health():
     return {"status": "ok"}
 @app.post("/run_tests", status_code=202)
 def run_tests():
     log("🚦 /run_tests çağrıldı. Testler başlatılıyor...")
 @app.get("/start", response_class=HTMLResponse)
 def root():
     session_id = str(uuid.uuid4())
     session_info = {
         "session_id": session_id,
         "last_intent": None,
         "awaiting_variable": None
     }
     if not hasattr(app.state, "session_store"):
         app.state.session_store = {}
     app.state.session_store[session_id] = session_info
     log(f"🌐 /start ile yeni session başlatıldı: {session_id}")
     return f"""
     <html><body>
         <h2>Turkcell LLM Chat</h2>
         <script>
         const sessionId = "{session_id}";
         localStorage.setItem("session_id", sessionId);
         async function send() {{
             const input = document.getElementById("input").value;
             const res = await fetch('/chat', {{
 def start_chat():
     if llm_model.model is None or llm_model.tokenizer is None:
         return {"error": "Model yüklenmedi."}
     if not hasattr(app.state, "session_store"):
         app.state.session_store = {}
     session_id = str(uuid.uuid4())
     session_info = {
         "session_id": session_id,
         if llm_model.model is None or llm_model.tokenizer is None:
             return {"error": "Model yüklenmedi."}
         detected_intent, intent_conf = None, 0.0
         if s_config.INTENT_MODEL:
             detected_intent, intent_conf = await intent.detect_intent(user_input)
         current_intent = session.get("last_intent")
         awaiting_variable = session.get("awaiting_variable")
         if (
             awaiting_variable and
             detected_intent and
             session["last_intent"] = detected_intent
             current_intent = detected_intent
         if (
             detected_intent and
             intent_conf > s_config.INTENT_CONFIDENCE_THRESHOLD and
             data_formats = s_config.DATA_FORMATS
             variable_format_map = definition.get("variable_formats", {})
             if awaiting_variable:
                 extracted = intent.extract_parameters(pattern_list, user_input)
                 for p in extracted:
                         log(f"✅ Awaiting parametre tamamlandı: {awaiting_variable} = {p['value']}")
                         break
             extracted = intent.extract_parameters(pattern_list, user_input)
             variables = {p["key"]: p["value"] for p in extracted}
             session.setdefault("variables", {}).update(variables)
             is_valid, validation_errors = intent.validate_variable_formats(session["variables"], variable_format_map, data_formats)
             if not is_valid:
                 session["awaiting_variable"] = list(validation_errors.keys())[0]
                 app.state.session_store = session_store
                 return {"response": list(validation_errors.values())[0]}
             expected_vars = list(variable_format_map.keys())
             missing_vars = [v for v in expected_vars if v not in session["variables"]]
             if missing_vars:
                 app.state.session_store = session_store
                 return {"response": f"Lütfen {missing_vars[0]} bilgisini belirtir misiniz?"}
             result = intent.execute_intent(
                 detected_intent,
                 user_input,
             else:
                 return {"response": random.choice(s_config.FALLBACK_ANSWERS)}
         session["awaiting_variable"] = None
         session["variables"] = {}
         response, response_conf = await llm_model.generate_response(user_input, s_config)
         traceback.print_exc()
         return JSONResponse(content={"error": str(e)}, status_code=500)
+# 🧠 Model setup ve sunucu
 threading.Thread(target=llm_model.setup_model, kwargs={"s_config": s_config}, daemon=True).start()
 threading.Thread(target=lambda: uvicorn.run(app, host="0.0.0.0", port=7860), daemon=True).start()
+# 🧘 Eğitim sonrası uygulama restart olmasın diye bekleme
+log("⏸️ Eğitim tamamlandı. Servis bekleme modunda...")
 while True:
+    time.sleep(60)