Update fine_tune_inference_test.py

fine_tune_inference_test.py

@@ -2,18 +2,18 @@ import os
 import threading
 import uvicorn
 from fastapi import FastAPI, Request
-from fastapi.responses import HTMLResponse
+from fastapi.responses import HTMLResponse, JSONResponse
 from pydantic import BaseModel
 from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
 from datasets import load_dataset
-from fastapi.responses import JSONResponse
+from peft import PeftModel
 
 # ✅ Sabitler
 HF_TOKEN = os.environ.get("HF_TOKEN")
 MODEL_BASE = "UcsTurkey/kanarya-750m-fixed"
-FINE_TUNE_ZIP = "trained_model_000_100.zip"  # 👈 Değiştirilebilir
+FINE_TUNE_ZIP = "trained_model_000_100.zip"
 FINE_TUNE_REPO = "UcsTurkey/trained-zips"
-RAG_DATA_FILE = "merged_dataset_000_100.parquet" # 👈 Değiştirilebilir
+RAG_DATA_FILE = "merged_dataset_000_100.parquet"
 RAG_DATA_REPO = "UcsTurkey/turkish-general-culture-tokenized"
 
 # ✅ FastAPI app
@@ -85,9 +85,14 @@ def setup_model():
     with zipfile.ZipFile(zip_path, "r") as zip_ref:
         zip_ref.extractall(extract_dir)
 
-    print("🔁 Tokenizer ve model yükleniyor...")
+    print("🔁 Tokenizer yükleniyor...")
     tokenizer = AutoTokenizer.from_pretrained(os.path.join(extract_dir, "output"))
-    model = AutoModelForCausalLM.from_pretrained(os.path.join(extract_dir, "output"))
+
+    print("🧠 Base model indiriliyor...")
+    base_model = AutoModelForCausalLM.from_pretrained(MODEL_BASE, torch_dtype="auto")
+
+    print("➕ LoRA adapter uygulanıyor...")
+    model = PeftModel.from_pretrained(base_model, os.path.join(extract_dir, "output"))
 
     print("📚 RAG dataseti yükleniyor...")
     rag = load_dataset(RAG_DATA_REPO, data_files=RAG_DATA_FILE, split="train", token=HF_TOKEN)