fix build: default to lib instead of lib64

[eschmidbauer]

Description of changes:
Fix build in conda env & nvidia packages

cd FlexFlow
conda env create -f conda/environment.yml
conda activate flexflow
conda install -c "nvidia" \
  cuda \
  cuda-toolkit \
  cuda-nvcc \
  cudnn

mkdir build
cd build

cmake \
  -DCUDA_USE_STATIC_CUDA_RUNTIME=OFF \
  -DLegion_HIJACK_CUDART=OFF \
  -DINFERENCE_TESTS=OFF \
  -DLIBTORCH_PATH=~/libtorch \
  -DCMAKE_BUILD_TYPE=Release \
  -DFF_CUDA_ARCH=autodetect \
  -DCUDA_PATH=~/miniconda3/envs/flexflow \
  -DCUDNN_PATH=~/miniconda3/envs/flexflow \
  -DFF_USE_PYTHON=ON \
  -DFF_USE_NCCL=ON \
  -DNCCL_PATH=/usr/local/cuda \
  -DFF_BUILD_ALL_EXAMPLES=OFF \
  -DFF_BUILD_ALL_INFERENCE_EXAMPLES=ON \
  -DFF_USE_PREBUILT_LEGION=OFF \
  -DFF_USE_PREBUILT_NCCL=OFF \
  -DFF_USE_ALL_PREBUILT_LIBRARIES=OFF \
  -DFF_BUILD_UNIT_TESTS=OFF \
  -DFF_USE_AVX2=OFF \
  -DFF_MAX_DIM=5 \
  -DLEGION_MAX_RETURN_SIZE=262144 \
  -DROCM_PATH=/opt/rocm \
  -DFF_GPU_BACKEND=cuda ../config/../

make -j

---

This change is 

[goliaro]

Hi @eschmidbauer, could you let us know why this change is needed? It looks like some changes may be needed to make this work, so if I have a better understanding of what problem we're trying to solve, I can help! Thanks!

[eschmidbauer]

Hi @eschmidbauer, could you let us know why this change is needed? It looks like some changes may be needed to make this work, so if I have a better understanding of what problem we're trying to solve, I can help! Thanks!

I had to make this change in order to build it on ubuntu 22.04. Please disregard if there is another way to make this build on ubuntu

[jiazhihao]

Close this PR since we can successfully run FF without the path change.