You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository has been archived by the owner on Jun 29, 2023. It is now read-only.
Hi, I'm using Hygon DCU Z100 (It's a AMD-like GPU) in CentOS 7.6 with ROCm 4.0.1.
I tried to get this work for stanford alpaca extension for 8bit adam, but I came across problems:
make CUDA_VERSION=gfx1030 hip results in the following:
which: no nvcc in (/public/software/compiler/dtk-22.10.1/bin:/public/software/compiler/dtk-22.10.1/llvm/bin:/public/software/compiler/dtk-22.10.1/hip/bin:/public/software/compiler/dtk-22.10.1/hip/bin/hipify:/public/software/compiler/dtk-22.10.1/miopen/bin:/opt/hpc/software/mpi/hpcx/v2.11.0/gcc-7.3.1/bin:/opt/hpc/software/mpi/hpcx/v2.11.0/hcoll/bin:/opt/hpc/software/mpi/hpcx/v2.11.0/ucx_without_rocm/bin:/opt/rh/devtoolset-7/root/usr/bin:/opt/hpc/setfreq:/opt/gridview/slurm/bin:/opt/gridview/slurm/sbin:/opt/gridview/munge/bin:/opt/gridview/munge/sbin:/opt/clusconf/sbin:/opt/clusconf/bin:/work/home/ac842t8oeo/miniconda3/envs/alpaca/bin:/work/home/ac842t8oeo/miniconda3/condabin:/opt/hpc/setfreq:/usr/lib64/qt-3.3/bin:/work/home/ac842t8oeo/perl5/bin:/opt/gridview/slurm/bin:/opt/gridview/slurm/sbin:/opt/gridview/munge/bin:/opt/gridview/munge/sbin:/opt/clusconf/sbin:/opt/clusconf/bin:/usr/local/bin:/usr/bin:/usr/local/sbin:/usr/sbin:/work/home/ac842t8oeo/.local/bin:/work/home/ac842t8oeo/bin)
# /usr/bin/hipcc -std=c++14 -c -fPIC --amdgpu-target=gfx1030 -I /work/home/ac842t8oeo/bnb-zhr/bitsandbytes-rocm/csrc -I /work/home/ac842t8oeo/bnb-zhr/bitsandbytes-rocm/include -I /public/software/compiler/dtk-22.10.1/include -I/public/software/compiler/dtk-22.10.1/llvm/include -I/public/software/compiler/dtk-22.10.1/hip/include -I/public/software/compiler/dtk-22.10.1/miopen/include -I /opt/rocm/hipcub/include -o /work/home/ac842t8oeo/bnb-zhr/bitsandbytes-rocm/build/ops.o -D NO_CUBLASLT /work/home/ac842t8oeo/bnb-zhr/bitsandbytes-rocm/csrc/ops.cu
# /usr/bin/hipcc -std=c++14 -c -fPIC --amdgpu-target=gfx1030 -I /work/home/ac842t8oeo/bnb-zhr/bitsandbytes-rocm/csrc -I /work/home/ac842t8oeo/bnb-zhr/bitsandbytes-rocm/include -I /public/software/compiler/dtk-22.10.1/include -I/public/software/compiler/dtk-22.10.1/llvm/include -I/public/software/compiler/dtk-22.10.1/hip/include -I/public/software/compiler/dtk-22.10.1/miopen/include -I /opt/rocm/hipcub/include -o /work/home/ac842t8oeo/bnb-zhr/bitsandbytes-rocm/build/kernels.o -D NO_CUBLASLT /work/home/ac842t8oeo/bnb-zhr/bitsandbytes-rocm/csrc/kernels.cu
/public/software/compiler/dtk-22.10.1/bin/hipcc -std=c++14 -c -fPIC --offload-arch=gfx1030 -I /work/home/ac842t8oeo/bnb-zhr/bitsandbytes-rocm/csrc -I /work/home/ac842t8oeo/bnb-zhr/bitsandbytes-rocm/include -I /public/software/compiler/dtk-22.10.1/include -I/public/software/compiler/dtk-22.10.1/llvm/include -I/public/software/compiler/dtk-22.10.1/hip/include -I/public/software/compiler/dtk-22.10.1/miopen/include -I /opt/rocm/hipcub/include -o /work/home/ac842t8oeo/bnb-zhr/bitsandbytes-rocm/build/ops.o -D NO_CUBLASLT /work/home/ac842t8oeo/bnb-zhr/bitsandbytes-rocm/csrc/ops.cu
/public/software/compiler/dtk-22.10.1/bin/hipcc -std=c++14 -c -fPIC --offload-arch=gfx1030 -I /work/home/ac842t8oeo/bnb-zhr/bitsandbytes-rocm/csrc -I /work/home/ac842t8oeo/bnb-zhr/bitsandbytes-rocm/include -I /public/software/compiler/dtk-22.10.1/include -I/public/software/compiler/dtk-22.10.1/llvm/include -I/public/software/compiler/dtk-22.10.1/hip/include -I/public/software/compiler/dtk-22.10.1/miopen/include -I /opt/rocm/hipcub/include -o /work/home/ac842t8oeo/bnb-zhr/bitsandbytes-rocm/build/kernels.o -D NO_CUBLASLT /work/home/ac842t8oeo/bnb-zhr/bitsandbytes-rocm/csrc/kernels.cu
/work/home/ac842t8oeo/bnb-zhr/bitsandbytes-rocm/csrc/kernels.cu:1874:40: error: no matching member function for call to 'BlockedToStriped'
BlockExchange(temp_storage.exchange).BlockedToStriped(local_col_absmax_values);
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~
/work/home/ac842t8oeo/bnb-zhr/bitsandbytes-rocm/csrc/kernels.cu:1899:26: note: in instantiation of function template specialization 'kgetColRowStats<__half, 64, 4, 16, 256, 0>' requested here
template __global__ void kgetColRowStats<half, 64, 4, 16, 64*4, 0>(half * __restrict__ A, float *rowStats, float *colStats, int * nnz_count_row, float nnz_threshold, int rows, int cols, int tiledRows, int tiledCols);
/opt/rocm/hipcub/include/hipcub/rocprim/block/block_exchange.hpp:91:10: note: candidate function template not viable: requires 2 arguments, but 1 was provided
void BlockedToStriped(InputT (&input_items)[ITEMS_PER_THREAD],
/work/home/ac842t8oeo/bnb-zhr/bitsandbytes-rocm/csrc/kernels.cu:1874:40: error: no matching member function for call to 'BlockedToStriped'
BlockExchange(temp_storage.exchange).BlockedToStriped(local_col_absmax_values);
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~
/work/home/ac842t8oeo/bnb-zhr/bitsandbytes-rocm/csrc/kernels.cu:1900:26: note: in instantiation of function template specialization 'kgetColRowStats<__half, 64, 4, 16, 256, 1>' requested here
template __global__ void kgetColRowStats<half, 64, 4, 16, 64*4, 1%3E(half * __restrict__ A, float *rowStats, float *colStats, int * nnz_count_row, float nnz_threshold, int rows, int cols, int tiledRows, int tiledCols);
/opt/rocm/hipcub/include/hipcub/rocprim/block/block_exchange.hpp:91:10: note: candidate function template not viable: requires 2 arguments, but 1 was provided
void BlockedToStriped(InputT (&input_items)[ITEMS_PER_THREAD],
2 errors generated when compiling for gfx1030.
make: *** [Makefile:113: hip] Error 1
Any assistance in this matter would be much appreciated, and thanks for your time!
The text was updated successfully, but these errors were encountered:
Sign up for freeto subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Hi, I'm using Hygon DCU Z100 (It's a AMD-like GPU) in CentOS 7.6 with ROCm 4.0.1.
I tried to get this work for stanford alpaca extension for 8bit adam, but I came across problems:
make CUDA_VERSION=gfx1030 hip
results in the following:Any assistance in this matter would be much appreciated, and thanks for your time!
The text was updated successfully, but these errors were encountered: