Update modules/pmbl.py

modules/pmbl.py

@@ -102,7 +102,7 @@ class PMBL:
             yield chunk
 
     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, use_flash_attn=True, mlock=True)
+        llm = Llama(model_path=self.model_path, n_ctx=n_ctx, n_threads=8, n_gpu_layers=-1, offload_kqv=True, flash_attn=True, use_mlock=True)
 
         response = llm(
             system_prompt,
@@ -147,7 +147,7 @@ class PMBL:
         conn.close()
 
     def generate_topic(self, prompt, response):
-        llm = Llama(model_path=self.model_path, n_ctx=1690, n_threads=2, n_gpu_layers=-1, mlock=True)
+        llm = Llama(model_path=self.model_path, n_ctx=n_ctx, n_threads=4, n_gpu_layers=-1, offload_kqv=True, flash_attn=True, use_mlock=True)
 
         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:"