pyc • Lines: 782�
�]�j�Q � �� � d Z ddlZddlZddlZddlZddlmZ ddd�Zdd�Zdd�Z dd �Z
dd
�Zdd�Zd� Z
dd
edefd�Zd� Zd� Zd� Zd� Zd� Zd� Zdd�Zdd�Zdd�Zdd�Zd� Zdd�Zd� Zy) a� llama-server lifecycle and the streaming LLM round-trip worker.
Two llama-server processes run concurrently:
* the **GPU** server on ``LLAMA_BASE`` (8081) serving interactive chat UI
users, and
* the **CPU** server on ``LLAMA_BASE_CPU`` (8079) serving automated self-chat
agents.
Every function below takes a ``mode`` (``"gpu"`` or ``"cpu"``) so loads,
unloads and completions always hit the right server without ever stopping the
other one.
� N)�Mzgpu_slot0.kvzcpu_slot0.kv��gpu�cpuc �<