Revert

modules/pmbl.py

@@ -102,18 +102,9 @@ class PMBL:
             yield chunk
 
     def generate_response_task(self, system_prompt, prompt, n_ctx):
-        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_mlock=True,
-                use_mmap=True,
-                use_flash_attn=True
-            )
-
-        response = llm.generate(
+        llm = Llama(model_path=self.model_path, n_ctx=n_ctx, n_threads=8, n_gpu_layers=-1, use_flash_attn=True, mlock=True)
+
+        response = llm(
             system_prompt,
             max_tokens=1500,
             temperature=0.7,
@@ -156,12 +147,7 @@ class PMBL:
         conn.close()
 
     def generate_topic(self, prompt, response):
-            llm = Llama(
-                model_path=self.model_path,
-                n_ctx=n_ctx,
-                n_threads=2,
-                n_gpu_layers=2,
-            )
+        llm = Llama(model_path=self.model_path, n_ctx=1690, n_threads=2, n_gpu_layers=-1, 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:"