use cpu temporarily

modules/pmbl.py

@@ -106,7 +106,7 @@ class PMBL:
             
     
     def generate_response_task(self, system_prompt, prompt, n_ctx):
-        llm = Llama(model_path=self.model_path, n_ctx=n_ctx, n_threads=8, n_gpu_layers=-1, mlock=True)
+        llm = Llama(model_path=self.model_path, n_ctx=n_ctx, n_threads=8, mlock=True)
         llm = llm.to("cuda")  # Move the model to the GPU
 
         response = llm(
@@ -154,7 +154,7 @@ class PMBL:
         
     @spaces.GPU
     def generate_topic(self, prompt, response):
-        llm = Llama(model_path=self.model_path, n_ctx=1690, n_threads=8, n_gpu_layers=-1, mlock=True)
+        llm = Llama(model_path=self.model_path, n_ctx=1690, n_threads=8, mlock=True)
         llm = llm.to("cuda")  # Move the model to the GPU
 
         system_prompt = f"Based on the following interaction between a user and an AI assistant, generate a concise topic for the conversation in 2-4 words:\n\nUser: {prompt}\nAssistant: {response}\n\nTopic:"