diff --git a/install.py b/install.py
index 6f2c8ed..a3e92d8 100644
--- a/install.py
+++ b/install.py
@@ -20,13 +20,14 @@ pip_list_lines = pip_list_str.splitlines()
 cuda_version = [item for item in pip_list_lines if item.startswith('torch')][0].split()[-1].split('+cu')[-1]
 llama_cpp_versions = [item for item in pip_list_lines if item.startswith('llama_cpp_python')]
 if len(llama_cpp_versions) > 0:
-    llama_cpp_cuda_version = llama_cpp_versions[0].split()[-1].split('+cu')[-1]
-if len(llama_cpp_versions) <= 0 or cuda_version != llama_cpp_cuda_version:
+    llama_cpp_version_splited = llama_cpp_versions[0].split()[-1].split('+cu')
+    llama_cpp_cuda_version = llama_cpp_version_splited[-1]
+if len(llama_cpp_versions) <= 0 or cuda_version != llama_cpp_cuda_version or llama_cpp_version_splited[0] != '0.2.36':
     import os
     if os.name == 'nt':
-        launch.run_pip('install https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/wheels/llama_cpp_python-0.2.23+cu' + cuda_version + '-cp310-cp310-win_amd64.whl', 'llama-cpp-python')
+        launch.run_pip('install https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/wheels/llama_cpp_python-0.2.36+cu' + cuda_version + '-cp310-cp310-win_amd64.whl', 'llama-cpp-python')
     else:
-        launch.run_pip('install https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/wheels/llama_cpp_python-0.2.23+cu' + cuda_version + '-cp310-cp310-manylinux_2_31_x86_64.whl', 'llama-cpp-python')
+        launch.run_pip('install https://github.com/oobabooga/llama-cpp-python-cuBLAS-wheels/releases/download/wheels/llama_cpp_python-0.2.36+cu' + cuda_version + '-cp310-cp310-manylinux_2_31_x86_64.whl', 'llama-cpp-python')
 
 if not launch.is_installed('gpt-stream-json-parser'):
     launch.run_pip('install git+https://github.com/furnqse/gpt-stream-json-parser.git', 'gpt-stream-json-parser')