Update app.py

app.py

@@ -1,33 +1,52 @@
 import os
-os.system("pip install --force-reinstall --no-cache-dir https://github.com/abetlen/llama-cpp-python/releases/download/v0.2.90/llama_cpp_python-0.2.90-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl")
+import subprocess
+import sys
 
+# --- ZONA MAGICĂ DE INSTALARE ---
+# Verificăm dacă llama_cpp e instalat. Dacă nu, îl instalăm acum, la pornire.
+try:
+    import llama_cpp
+    print("✅ Llama-cpp-python este deja instalat.")
+except ImportError:
+    print("⏳ Se instalează llama-cpp-python pentru CPU (poate dura 30-60 secunde)...")
+    # Această comandă folosește index-ul oficial abetlen, care găsește singur versiunea corectă
+    subprocess.check_call([
+        sys.executable, "-m", "pip", "install", 
+        "llama-cpp-python", 
+        "--no-cache-dir", 
+        "--extra-index-url", "https://abetlen.github.io/llama-cpp-python/whl/cpu"
+    ])
+    print("✅ Instalare completă!")
+
+# --- ACUM IMPORTĂM BIBLIOTECILE ---
 import gradio as gr
 from llama_cpp import Llama
 from huggingface_hub import hf_hub_download
-REPO_ID = "i04n4/llama3.2-3b-math-gguf"
-FILENAME = "model.gguf"
 
-model_path = hf_hub_download(repo_id=REPO_ID, filename=FILENAME)
+# --- CONFIGURARE ---
+REPO_ID = "IoanaUser/Llama-3.2-3B-Math-GGUF" # <--- Verifică să fie numele tău corect
+FILENAME = "llama-3.2-3b-math-q4_k_m.gguf"   # <--- Verifică numele fișierului
+
+print(f"⏳ Descarc modelul {FILENAME}...")
+try:
+    model_path = hf_hub_download(repo_id=REPO_ID, filename=FILENAME)
+except Exception as e:
+    print(f"❌ Eroare la descărcare: {e}")
+    raise e
 
 print("🚀 Încarc motorul Llama...")
+# n_threads=2 este optim pentru Free Tier
 llm = Llama(model_path=model_path, n_ctx=2048, n_threads=2)
 
 def generate_response(message, history):
-    system_prompt = """You are a specialized mathematical assistant. Your goal is to be precise and efficient.
-
-RULES:
-- RESPONSE TYPE A (Basic/Fact-based): If the question is simple arithmetic (e.g., 2+2) or a basic fact, provide ONLY the final result. Be as concise as possible.
-- RESPONSE TYPE B (Complex/Logic): If the question requires multiple steps, algebra, or reasoning, provide a clear 'Step-by-step reasoning' followed by 'Final Answer'.
-- FORMATTING: Never use internal scratchpad tags like <<...>>. Use plain text only. Do not use your inner thougths, only the explanation. Do not use any symbols that are not needed
-- LANGUAGE: Always answer in English."""
-    
+    system_prompt = "You are a helpful mathematical assistant. Answer directly and precisely."
     full_prompt = f"<|start_header_id|>system<|end_header_id|>\n\n{system_prompt}<|eot_id|><|start_header_id|>user<|end_header_id|>\n\n{message}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n"
     
     response = ""
     output = llm(
         full_prompt, 
         max_tokens=512, 
-        stop=["<|eot_id|>", "User:"], 
+        stop=["<|eot_id|>"], 
         echo=False, 
         stream=True
     )
@@ -39,9 +58,8 @@ RULES:
 
 demo = gr.ChatInterface(
     generate_response, 
-    title="Super Awesome Math Tutor",
-    description="Ask me a math question!!!!!!",
-    examples=["Calculate 25 * 14", "Solve 2x + 5 = 15", "Derivative of x^2"]
+    title="Math tutor",
+    description="Ask me a question vro<3",
 )
 
 if __name__ == "__main__":