Update app.py

app.py

@@ -1,44 +1,34 @@
-import torch
-from transformers import AutoTokenizer, AutoModelForCausalLM
 import gradio as gr
 import json
-import random
+from sentence_transformers import SentenceTransformer
+from sklearn.metrics.pairwise import cosine_similarity
+import numpy as np
 
-# Model yükle
-model_id = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
-tokenizer = AutoTokenizer.from_pretrained(model_id)
-model = AutoModelForCausalLM.from_pretrained(model_id)
+# 1. Küçük bir gömme (embedding) modeli
+embedder = SentenceTransformer("all-MiniLM-L6-v2")
 
-# Hafızadan örnekleri yükle
+# 2. Verileri yükle
 with open("memory_questions.json", "r") as f:
     memory_data = json.load(f)
 
-# Few-shot prompt oluşturan fonksiyon
-def get_few_shot_prompt(memory, k=5):
-    examples = random.sample(memory_data, k)
-    few_shot = "\n".join(
-        [f"Memory: {ex['description']}\nQuestion: {ex['question']}" for ex in examples]
-    )
-    return f"{few_shot}\nMemory: {memory}\nQuestion:"
+# 3. Tüm memory'leri embed et (başta bir defa)
+memory_texts = [item['description'] for item in memory_data]
+memory_embeddings = embedder.encode(memory_texts)
 
-# Ana fonksiyon
-def generate_question(memory):
-    prompt = get_few_shot_prompt(memory)
-    input_ids = tokenizer(prompt, return_tensors="pt").input_ids
-    output = model.generate(input_ids, max_new_tokens=50, do_sample=False)
-    result = tokenizer.decode(output[0], skip_special_tokens=True)
+# 4. Soru üretme fonksiyonu
+def generate_question(user_memory):
+    user_embedding = embedder.encode([user_memory])
+    similarities = cosine_similarity(user_embedding, memory_embeddings)[0]
+    best_match_index = np.argmax(similarities)
+    return memory_data[best_match_index]['question']
 
-    # Sadece cevabı ayıkla
-    lines = result.strip().split("Question:")
-    return lines[-1].strip() if len(lines) > 1 else result.strip()
-
-# Gradio arayüzü
+# 5. Gradio Arayüzü
 iface = gr.Interface(
     fn=generate_question,
     inputs=gr.Textbox(label="Your Memory"),
     outputs=gr.Textbox(label="Generated Question"),
-    title="MemoRease - TinyLLaMA Chat Generator",
-    description="Write a memory, get a question to help recall more details!"
+    title="MemoRease - Semantic Memory Question Generator",
+    description="Write a memory, get the most semantically related question from your dataset!"
 )
 
 iface.launch()