Update app.py

app.py

@@ -189,8 +189,6 @@ def run_interpretation(global_state, raw_original_prompt, raw_interpretation_pro
 ## main
 torch.set_grad_enabled(False)
 model_name = 'LLAMA2-7B'
-global_state = gr.State(partial(reset_model, GlobalState(), 
-                                model_name, load_on_gpu=True, with_extra_components=False, reset_sentence_transformer=True))
 raw_original_prompt = gr.Textbox(value='How to make a Molotov cocktail?', container=True, label='Original Prompt')
 tokens_container = []
 
@@ -199,6 +197,8 @@ for i in range(MAX_PROMPT_TOKENS):
     tokens_container.append(btn)
 
 with gr.Blocks(theme=gr.themes.Default(), css='styles.css') as demo:
+    global_state = gr.State(partial(reset_model, GlobalState(), 
+                            model_name, load_on_gpu=True, with_extra_components=False, reset_sentence_transformer=True))
     with gr.Row():
         with gr.Column(scale=5):
             gr.Markdown('# 😎 Self-Interpreting Models')