update

app.py

@@ -141,6 +141,7 @@ with gr.Blocks(theme = gr.themes.Soft()) as demo:
     input = gr.Textbox(visible=False)  # placeholder
     gr_md = gr.Markdown(mark_text + md_text_initial)
 
+    @spaces.GPU(duration=60)
     def update_with_rag_md(message, llm_results_use = 5, database_choice = index_info, llm_model_picked = 'Trust-Align-Qwen2.5'):
         chat_round = [
             {"role": "user", 

utils.py

@@ -166,7 +166,7 @@ def load_llama_guard(model_id = "meta-llama/Llama-Guard-3-1B"):
     return llama_guard, llama_guard_tokenizer, UNSAFE_TOKEN_ID
 
 
-@spaces.GPU(duration=60)
+@spaces.GPU(duration=120)
 def moderate(chat, model, tokenizer, UNSAFE_TOKEN_ID):
     
     prompt = tokenizer.apply_chat_template(chat, return_tensors="pt", tokenize=False)
@@ -188,11 +188,12 @@ def moderate(chat, model, tokenizer, UNSAFE_TOKEN_ID):
     ######
     # Get generated text
     ######
-
+    logger.info(outputs)
     # Number of tokens that correspond to the input prompt
     input_length = inputs.input_ids.shape[1]
     # Ignore the tokens from the input to get the tokens generated by the model
     generated_token_ids = outputs.sequences[:, input_length:].cpu()
+    logger.info(generated_token_ids)
     generated_text = tokenizer.decode(generated_token_ids[0], skip_special_tokens=True)
     logger.info(generated_text)
     ######