use more thread for inference

app.py

@@ -91,7 +91,7 @@ def update_llm(size, model_file, clip_file):
     if (model_cache['size'], model_cache['model_file'], model_cache['clip_file']) != (size, model_file, clip_file):
         mf, cf = ensure_weights(size, model_file, clip_file)
         handler = SmolVLM2ChatHandler(clip_model_path=cf, verbose=False)
-        llm = Llama(model_path=mf, chat_handler=handler, n_ctx=1024, verbose=False)
+        llm = Llama(model_path=mf, chat_handler=handler, n_ctx=1024, verbose=False, n_threads=min(2, os.cpu_count()), )
         model_cache.update({'size': size, 'model_file': mf, 'clip_file': cf, 'llm': llm})
     return None  # no UI output