update

app.py

@@ -35,7 +35,7 @@ generate_kwargs = dict(
 )
 
 # Load llama Guard
-llama_guard, llama_guard_tokenizer, UNSAFE_TOKEN_ID = load_llama_guard("meta-llama/Llama-Guard-3-1B")
+# llama_guard, llama_guard_tokenizer, UNSAFE_TOKEN_ID = load_llama_guard("meta-llama/Llama-Guard-3-1B")
 
 ## RAG MODEL
 RAG = RAGPretrainedModel.from_index("colbert/indexes/arxiv_colbert", n_gpu=0)
@@ -143,17 +143,17 @@ with gr.Blocks(theme = gr.themes.Soft()) as demo:
 
     # @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", 
-            "content": [
-                {"type": "text", 
-                "text": message
-                }
-                ]
-            }
-        ]
+        # chat_round = [
+        #     {"role": "user", 
+        #     "content": [
+        #         {"type": "text", 
+        #         "text": message
+        #         }
+        #         ]
+        #     }
+        # ]
         # llama guard check for it
-        prompt_safety = moderate(chat_round, llama_guard, llama_guard_tokenizer, UNSAFE_TOKEN_ID)['generated_text']
+        # prompt_safety = moderate(chat_round, llama_guard, llama_guard_tokenizer, UNSAFE_TOKEN_ID)['generated_text']
         prompt_safety = "safe"
         
         if prompt_safety == "safe":

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=120)
+# @spaces.GPU(duration=120)
 def moderate(chat, model, tokenizer, UNSAFE_TOKEN_ID):
     
     prompt = tokenizer.apply_chat_template(chat, return_tensors="pt", tokenize=False)