Refine config and runtime flow

src/aiprocess.py

@@ -5,12 +5,9 @@ import logging
 import os
 import sys
 import urllib.request
-from dataclasses import dataclass
 from typing import Any, Callable, cast
 
-from llama_cpp import Llama, llama_cpp as llama_cpp_lib  # type: ignore[import-not-found]
-
-from constants import LLM_LANGUAGE, MODEL_DIR, MODEL_NAME, MODEL_PATH, MODEL_URL
+from constants import MODEL_DIR, MODEL_NAME, MODEL_PATH, MODEL_URL
 
 
 SYSTEM_PROMPT = (
@@ -36,7 +33,8 @@ SYSTEM_PROMPT = (
 
 
 class LlamaProcessor:
-    def __init__(self, verbose=False):
+    def __init__(self, verbose: bool = False):
+        Llama, llama_cpp_lib = _load_llama_bindings()
         ensure_model()
         if not verbose:
             os.environ.setdefault("LLAMA_CPP_LOG_LEVEL", "ERROR")
@@ -100,6 +98,16 @@ def ensure_model():
         raise
 
 
+def _load_llama_bindings():
+    try:
+        from llama_cpp import Llama, llama_cpp as llama_cpp_lib  # type: ignore[import-not-found]
+    except ModuleNotFoundError as exc:
+        raise RuntimeError(
+            "llama-cpp-python is not installed; install dependencies with `uv sync`"
+        ) from exc
+    return Llama, llama_cpp_lib
+
+
 def _extract_chat_text(payload: Any) -> str:
     if "choices" in payload and payload["choices"]:
         choice = payload["choices"][0]