Prevent llama.cpp from being monkey patched more than once (closes oobabooga#6201)

Author: oobabooga

modules/llama_cpp_python_hijack.py

@@ -100,9 +100,11 @@ def eval_with_progress(self, tokens: Sequence[int]):
 
 
 def monkey_patch_llama_cpp_python(lib):
+    if getattr(lib.Llama, '_is_patched', False):
+        # If the patch is already applied, do nothing
+        return
 
     def my_generate(self, *args, **kwargs):
-
         if shared.args.streaming_llm:
             new_sequence = args[0]
             past_sequence = self._input_ids
@@ -116,3 +118,6 @@ def my_generate(self, *args, **kwargs):
     lib.Llama.eval = eval_with_progress
     lib.Llama.original_generate = lib.Llama.generate
     lib.Llama.generate = my_generate
+
+    # Set the flag to indicate that the patch has been applied
+    lib.Llama._is_patched = True