fix cuda pypi package (#2423)

* fix cuda pypi package

* patch bump

mlx/version.h

@@ -3,8 +3,8 @@
 #pragma once
 
 #define MLX_VERSION_MAJOR 0
-#define MLX_VERSION_MINOR 26
+#define MLX_VERSION_MINOR 27
-#define MLX_VERSION_PATCH 5
+#define MLX_VERSION_PATCH 1
 #define MLX_VERSION_NUMERIC \
   (100000 * MLX_VERSION_MAJOR + 1000 * MLX_VERSION_MINOR + MLX_VERSION_PATCH)
 

python/scripts/repair_cuda.sh

@@ -5,6 +5,7 @@ auditwheel repair dist/* \
   --exclude libcublas* \
   --exclude libnvrtc* \
   --exclude libcuda* \
+  --exclude libcudnn* \
   -w wheel_tmp
 
 

setup.py

@@ -289,7 +289,7 @@ if __name__ == "__main__":
             install_requires += [
                 "nvidia-cublas-cu12==12.9.*",
                 "nvidia-cuda-nvrtc-cu12==12.9.*",
-                "nvidia-cudnn-cu12==12.9.*",
+                "nvidia-cudnn-cu12==9.*",
             ]
         else:
             name = "mlx-cpu"