Update app.py

app.py

@@ -3,59 +3,61 @@ from transformers import AutoModelForCausalLM, AutoTokenizer, TrainingArguments,
 from peft import LoraConfig, get_peft_model
 from datasets import load_dataset
 import gradio as gr
-
-# === 1️⃣ MODEL VE TOKENIZER YÜKLEME ===
-MODEL_NAME = "mistralai/Mistral-7B-v0.1"  # Hugging Face model adı
-tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
-
-# === 2️⃣ CPU OPTİMİZASYONU ===
-torch_dtype = torch.float32  # CPU için uygun dtype
-model = AutoModelForCausalLM.from_pretrained(MODEL_NAME, torch_dtype=torch_dtype)
-
-# === 3️⃣ LoRA AYARLARI ===
-lora_config = LoraConfig(
-    r=8,              
-    lora_alpha=32,    
-    lora_dropout=0.1, 
-    bias="none",
-    target_modules=["q_proj", "v_proj"], 
-)
-model = get_peft_model(model, lora_config)
-
-# === 4️⃣ VERİ SETİ ===
-dataset = load_dataset("oscar", "unshuffled_deduplicated_tr", trust_remote_code=True)  # trust_remote_code=True
-subset = dataset["train"].shuffle(seed=42).select(range(10000))  # Küçük subset seçiyoruz (10.000 örnek)
-
-# === 5️⃣ TOKENLEŞTİRME FONKSİYONU ===
-@spaces.GPU
-def tokenize_function(examples):
-    return tokenizer(examples["text"], truncation=True, max_length=512)
-
-tokenized_datasets = subset.map(tokenize_function, batched=True)
-
-# === 6️⃣ EĞİTİM AYARLARI ===
-# Eğitimde kaç adım olduğunu hesaplayalım
-train_size = len(tokenized_datasets)  # 10,000 örnek
-batch_size = 1  # Batch size 1
-num_epochs = 1  # 1 epoch eğitimi
-max_steps = (train_size // batch_size) * num_epochs  # max_steps hesapla
-
-training_args = TrainingArguments(
-    output_dir="./mistral_lora",
-    per_device_train_batch_size=1,  
-    gradient_accumulation_steps=16, 
-    learning_rate=5e-4,  
-    num_train_epochs=1,  
-    max_steps=max_steps,  # Buraya max_steps parametresini ekliyoruz
-    save_steps=500,
-    save_total_limit=2,
-    logging_dir="./logs",
-    logging_steps=10,
-    optim="adamw_torch",
-    train_dataset=split_dataset["train"],
-    eval_dataset=split_dataset["test"],
-    no_cuda=True,  # GPU kullanılmıyor
-)
+    @spaces.GPU
+    def main():
+    # === 1️⃣ MODEL VE TOKENIZER YÜKLEME ===
+    MODEL_NAME = "mistralai/Mistral-7B-v0.1"  # Hugging Face model adı
+    tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
+    
+    # === 2️⃣ CPU OPTİMİZASYONU ===
+    torch_dtype = torch.float32  # CPU için uygun dtype
+    model = AutoModelForCausalLM.from_pretrained(MODEL_NAME, torch_dtype=torch_dtype)
+    
+    # === 3️⃣ LoRA AYARLARI ===
+    lora_config = LoraConfig(
+        r=8,              
+        lora_alpha=32,    
+        lora_dropout=0.1, 
+        bias="none",
+        target_modules=["q_proj", "v_proj"], 
+    )
+    model = get_peft_model(model, lora_config)
+    
+    # === 4️⃣ VERİ SETİ ===
+    dataset = load_dataset("oscar", "unshuffled_deduplicated_tr", trust_remote_code=True)  # trust_remote_code=True
+    subset = dataset["train"].shuffle(seed=42).select(range(10000))  # Küçük subset seçiyoruz (10.000 örnek)
+    
+    # === 5️⃣ TOKENLEŞTİRME FONKSİYONU ===
+    @spaces.GPU
+    def tokenize_function(examples):
+        return tokenizer(examples["text"], truncation=True, max_length=512)
+    
+    tokenized_datasets = subset.map(tokenize_function, batched=True)
+    
+    # === 6️⃣ EĞİTİM AYARLARI ===
+    # Eğitimde kaç adım olduğunu hesaplayalım
+    train_size = len(tokenized_datasets)  # 10,000 örnek
+    batch_size = 1  # Batch size 1
+    num_epochs = 1  # 1 epoch eğitimi
+    max_steps = (train_size // batch_size) * num_epochs  # max_steps hesapla
+    
+    training_args = TrainingArguments(
+        output_dir="./mistral_lora",
+        per_device_train_batch_size=1,  
+        gradient_accumulation_steps=16, 
+        learning_rate=5e-4,  
+        num_train_epochs=1,  
+        max_steps=max_steps,  # Buraya max_steps parametresini ekliyoruz
+        save_steps=500,
+        save_total_limit=2,
+        logging_dir="./logs",
+        logging_steps=10,
+        optim="adamw_torch",
+        train_dataset=split_dataset["train"],
+        eval_dataset=split_dataset["test"],
+        no_cuda=True,  # GPU kullanılmıyor
+    )
+main()
 
 # === 7️⃣ MODEL EĞİTİMİ ===
 @spaces.GPU