From aa653e3b5af8a9a977c75cd55f4cbc843405da7e Mon Sep 17 00:00:00 2001
From: oobabooga <112222186+oobabooga@users.noreply.github.com>
Date: Fri, 5 Jul 2024 03:34:15 -0700
Subject: [PATCH] Prevent llama.cpp from being monkey patched more than once
 (closes #6201)

---
 modules/llama_cpp_python_hijack.py | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/modules/llama_cpp_python_hijack.py b/modules/llama_cpp_python_hijack.py
index f3f3f560..c9813708 100644
--- a/modules/llama_cpp_python_hijack.py
+++ b/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 monkey_patch_llama_cpp_python(lib):
     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