Update app.py

app.py

@@ -1,30 +1,40 @@
 import torch
 from transformers import AutoTokenizer, AutoModelForCausalLM
 import gradio as gr
+import json
 
-# Model yükleme (hafif!)
 model_id = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
-
 tokenizer = AutoTokenizer.from_pretrained(model_id)
 model = AutoModelForCausalLM.from_pretrained(model_id)
 
+# Örnek veri setini yükle
+with open("memory_questions.json", "r") as f:
+    memory_data = json.load(f)
+
+# İlk 3-4 örnekten prompt hazırla
+few_shot_examples = "\n".join(
+    [f"Memory: {item['description']}\nQuestion: {item['question']}" for item in memory_data[:5]]
+)
+
 def generate_question(memory):
-    prompt = f"<|system|>You are a helpful assistant that generates meaningful questions from memories.<|user|>Memory: {memory}\nGenerate a related question.<|assistant|>"
+    prompt = f"""{few_shot_examples}
+Memory: {memory}
+Question:"""
 
     input_ids = tokenizer(prompt, return_tensors="pt").input_ids
-    output_ids = model.generate(input_ids, max_new_tokens=50, do_sample=True)
-    result = tokenizer.decode(output_ids[0], skip_special_tokens=True)
+    output = model.generate(input_ids, max_new_tokens=50, do_sample=False)
+    result = tokenizer.decode(output[0], skip_special_tokens=True)
 
-    # Cevabın son kısmını al
-    question = result.split("<|assistant|>")[-1].strip()
-    return question
+    # Çıktıdan yalnızca son soruyu ayrıştır
+    lines = result.strip().split("Question:")
+    return lines[-1].strip() if len(lines) > 1 else result.strip()
 
 # Gradio UI
 iface = gr.Interface(
     fn=generate_question,
     inputs=gr.Textbox(label="Your Memory"),
     outputs=gr.Textbox(label="Generated Question"),
-    title="TinyLLaMA Memory Question Generator"
+    title="Memory-Aware Question Generator (TinyLLaMA)"
 )
 
 iface.launch()