Attempting to compile pytorch 1.13 from source witrh CUDA 11.4 on Xavier NX
[40/593] Slimming NCCL
FAILED: nccl/lib/libnccl_slim_static.a /mnt/sda1/pytorch/build/nccl/lib/libnccl_slim_static.a
cd /mnt/sda1/pytorch/build/nccl && /usr/bin/cmake -E make_directory /mnt/sda1/pytorch/build/nccl/objects && cd objects && /usr/bin/ar x /mnt/sda1/pytorch/build/nccl/lib/libnccl_static.a && for obj in all_gather_* all_reduce_* broadcast_* reduce_*.o; do /usr/bin/objcopy --remove-relocations .nvFatBinSegment --remove-section __nv_relfatbin $obj; done && /usr/bin/ar cr /mnt/sda1/pytorch/build/nccl/lib/libnccl_slim_static.a *.o && /usr/bin/ar xN 1 /mnt/sda1/pytorch/build/nccl/lib/libnccl_static.a net.o && /usr/bin/ar q /mnt/sda1/pytorch/build/nccl/lib/libnccl_slim_static.a net.o && cd - && /usr/bin/cmake -E remove_directory /mnt/sda1/pytorch/build/nccl/objects
/usr/bin/objcopy: stx6dsTj: symbol `__nv_relfatbin’ required but not present
/usr/bin/objcopy: stx6dsTj: no symbols
last error repeated hundreds of times
Any help appreciated
Thanks