Update fine_tune_inference_test_mistral.py

fine_tune_inference_test_mistral.py

@@ -1,4 +1,4 @@
-import os, torch, zipfile, threading, uvicorn
+import os, torch, zipfile, threading, uvicorn, time, traceback
 from fastapi import FastAPI
 from fastapi.responses import HTMLResponse, JSONResponse
 from pydantic import BaseModel
@@ -6,7 +6,6 @@ from transformers import AutoTokenizer, AutoModelForCausalLM
 from peft import PeftModel
 from huggingface_hub import hf_hub_download
 from datetime import datetime
-import time
 import random
 
 # === Sabitler ===
@@ -14,8 +13,8 @@ HF_TOKEN = os.environ.get("HF_TOKEN")
 MODEL_BASE = "mistralai/Mistral-7B-Instruct-v0.2"
 FINE_TUNE_ZIP = "trained_model_000_009.zip"
 FINE_TUNE_REPO = "UcsTurkey/trained-zips"
+USE_FINE_TUNE = False   # ✅ Ana modeli test etmek için False yap
 USE_SAMPLING = False
-USE_FINE_TUNE = False  # ✅ Ana model ile test için False yap
 CONFIDENCE_THRESHOLD = -1.5
 FALLBACK_ANSWERS = [
     "Bu konuda maalesef bilgim yok.",
@@ -71,13 +70,16 @@ def root():
 def chat(msg: Message):
     global model, tokenizer
     try:
-        if model is None:
-            return {"error": "Model yüklenmedi"}
+        if model is None or tokenizer is None:
+            return {"error": "Model veya tokenizer henüz yüklenmedi."}
+
         user_input = msg.user_input.strip()
         if not user_input:
             return {"error": "Boş giriş"}
+
         prompt = f"SORU: {user_input}\nCEVAP:"
         inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
+
         with torch.no_grad():
             output = model.generate(
                 **inputs,
@@ -90,6 +92,7 @@ def chat(msg: Message):
                 output_scores=True,
                 suppress_tokens=[tokenizer.pad_token_id]
             )
+
         decoded = tokenizer.decode(output.sequences[0], skip_special_tokens=True)
         answer = decoded[len(prompt):].strip()
 
@@ -106,8 +109,10 @@ def chat(msg: Message):
         chat_history.append({"user": user_input, "bot": answer})
         log(f"Soru: {user_input} → Cevap: {answer[:60]}...")
         return {"answer": answer, "chat_history": chat_history}
+
     except Exception as e:
         log(f"❌ /chat hatası: {e}")
+        traceback.print_exc()
         return {"error": str(e)}
 
 def detect_env():
@@ -118,44 +123,52 @@ def detect_env():
 def setup_model():
     global model, tokenizer
     try:
-        extract_path = "/app/extracted"
         device, supports_bf16 = detect_env()
         dtype = torch.bfloat16 if supports_bf16 else torch.float32
         log(f"🧠 Ortam: {device.upper()}, dtype: {dtype}")
 
         if USE_FINE_TUNE:
-            log("📦 Zip indiriliyor...")
+            log("📦 Fine-tune zip indiriliyor...")
             zip_path = hf_hub_download(
                 repo_id=FINE_TUNE_REPO,
                 filename=FINE_TUNE_ZIP,
                 repo_type="model",
                 token=HF_TOKEN
             )
+            extract_path = "/app/extracted"
             os.makedirs(extract_path, exist_ok=True)
             with zipfile.ZipFile(zip_path, "r") as zip_ref:
                 zip_ref.extractall(extract_path)
+
             tokenizer = AutoTokenizer.from_pretrained(os.path.join(extract_path, "output"))
             if tokenizer.pad_token is None:
                 tokenizer.pad_token = tokenizer.eos_token
+
             base = AutoModelForCausalLM.from_pretrained(MODEL_BASE, torch_dtype=dtype).to(device)
             peft = PeftModel.from_pretrained(base, os.path.join(extract_path, "output"))
             model = peft.model.to(device)
+
         else:
-            log("🧪 Sadece ana model yüklenecek.")
+            log("🧪 Sadece ana model yüklenecek...")
             tokenizer = AutoTokenizer.from_pretrained(MODEL_BASE, use_fast=False)
+            if tokenizer.pad_token is None:
+                tokenizer.pad_token = tokenizer.eos_token
             model = AutoModelForCausalLM.from_pretrained(MODEL_BASE, torch_dtype=dtype).to(device)
 
         model.eval()
-        log("✅ Model yüklendi.")
+        log("✅ Model başarıyla yüklendi.")
+
     except Exception as e:
         log(f"❌ Model setup hatası: {e}")
+        traceback.print_exc()
 
 def run_server():
     log("🌐 Uvicorn başlatılıyor...")
     uvicorn.run(app, host="0.0.0.0", port=7860)
 
-log("🚀 Başlatılıyor...")
+log("===== Application Startup =====")
 threading.Thread(target=setup_model, daemon=True).start()
 threading.Thread(target=run_server, daemon=True).start()
+
 while True:
-    time.sleep(60)
+    time.sleep(60)