Update app.py

app.py

@@ -1,31 +1,31 @@
 import gradio as gr
-from transformers import AutoTokenizer, AutoModelForCausalLM
+from transformers import AutoTokenizer, GemmaForCausalLM
 from deep_translator import GoogleTranslator
 import torch
 
 # مشخصات مدل
-model_id = "google/gemma-3-4b-it"  # یا "google/gemma-2-7b-it" بسته به نیاز شما
+model_id = "google/gemma-3-4b-it"
 
 # بارگذاری مدل و توکنایزر
 tokenizer = AutoTokenizer.from_pretrained(model_id)
-model = AutoModelForCausalLM.from_pretrained(
+model = GemmaForCausalLM.from_pretrained(
     model_id,
-    torch_dtype=torch.bfloat16,
+    torch_dtype=torch.bfloat16 if torch.cuda.is_available() else torch.float32,
     device_map="auto"
 )
 model.eval()
 
 def generate_topics(field, major, keywords, audience, level):
     # ساخت پرامپت
-    prompt = f"""[INST]Suggest 3 academic thesis topics based on the following information:
+    prompt = f"""<bos>[INST]Suggest 3 academic thesis topics based on the following information:
 Field: {field}
 Specialization: {major}
 Keywords: {keywords}
 Target audience: {audience}
-Level: {level}[/INST]
-"""
+Level: {level}[/INST]"""
+
     # تولید خروجی
-    inputs = tokenizer(prompt, return_tensors="pt").to("cuda" if torch.cuda.is_available() else "cpu")
+    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
     with torch.no_grad():
         outputs = model.generate(**inputs, max_new_tokens=256)
     english_output = tokenizer.decode(outputs[0], skip_special_tokens=True)