Apple Silicon : mach-o file, but is an incompatible architecture (have 'x86_64', need 'arm64')

[mozzipa]

Because instruction for Apple silicon in Readme.md has been applied, CMAKE_SYSTEM_PROCESSOR from cmake . appears arm64. But, there is warning as below for cmake .

-- Accelerate framework found
-- CMAKE_SYSTEM_PROCESSOR: arm64
-- ARM detected
-- CMAKE_SYSTEM_PROCESSOR: arm64
CMake Warning at ggml/src/CMakeLists.txt:48 (message):
  Your arch is announced as x86_64, but it seems to actually be ARM64.  Not
  fixing that can lead to bad performance.  For more info see:
  https://github.com/ggerganov/whisper.cpp/issues/66#issuecomment-#1282546789


-- ARM detected
-- Accelerate framework found
-- Configuring done (0.1s)
-- Generating done (0.0s)
-- Build files have been written to: /my/path/rwkv.cpp

Furthermore, below error occurs for python rwkv/generate_completions.py /my/path/rwkv.cpp/models/rwkv-4-raven/ggml-model-q5_1.bin

OSError: dlopen(/my/path/rwkv.cpp/librwkv.dylib, 0x0006): tried: '/my/path/rwkv.cpp/librwkv.dylib' (mach-o file, but is an incompatible architecture (have 'x86_64', need 'arm64')), '/System/Volumes/Preboot/Cryptexes/OS/my/path/rwkv.cpp/librwkv.dylib' (no such file), '/my/path/rwkv.cpp/librwkv.dylib' (mach-o file, but is an incompatible architecture (have 'x86_64', need 'arm64'))

For your reference, similar error for llama-cpp-python was solved with arch -arm64 pip install llama-cpp-python --no-cache-dir It might be similar solution mentioned in above github link from warning.

Additionally, arch -arm64 cmake . and arch -arm64 cmake --build . --config Release were not valid.

[realcarlos]

same issues

[joakimeriksson]

I get similar problem even without that warning.

-- CMAKE_SYSTEM_PROCESSOR: arm64
-- ARM detected
-- Configuring done (0.0s)
-- Generating done (0.2s)
-- Build files have been written to: /Users/joakimeriksson/work/rwkv.cpp
[  9%] Built target ggml
[ 18%] Built target rwkv
[ 27%] Built target test_ggml_basics
[ 36%] Built target test_quantized_matmul_on_gpu
[ 45%] Built target test_tiny_rwkv
[ 54%] Built target test_quantization_format_compatibility
[ 63%] Built target test_logit_calculation_skipping
[ 72%] Built target test_eval_sequence_in_chunks
[ 81%] Built target test_context_cloning
[ 90%] Built target rwkv_cpu_info
[100%] Built target rwkv_quantize

Then:

python python/chat_with_bot.py ~/Downloads/rwkv.cpp-169M-Q5_1.bin
Traceback (most recent call last):
  File "/Users/joakimeriksson/work/rwkv.cpp/python/chat_with_bot.py", line 55, in <module>
    library = rwkv_cpp_shared_library.load_rwkv_shared_library()
  File "/Users/joakimeriksson/work/rwkv.cpp/python/rwkv_cpp/rwkv_cpp_shared_library.py", line 440, in load_rwkv_shared_library
    return RWKVSharedLibrary(str(full_path))
  File "/Users/joakimeriksson/work/rwkv.cpp/python/rwkv_cpp/rwkv_cpp_shared_library.py", line 45, in __init__
    self.library = ctypes.cdll.LoadLibrary(shared_library_path)
  File "/Users/joakimeriksson/anaconda3/envs/detic/lib/python3.9/ctypes/__init__.py", line 460, in LoadLibrary
    return self._dlltype(name)
  File "/Users/joakimeriksson/anaconda3/envs/detic/lib/python3.9/ctypes/__init__.py", line 382, in __init__
    self._handle = _dlopen(self._name, mode)
OSError: dlopen(/Users/joakimeriksson/work/rwkv.cpp/librwkv.dylib, 0x0006): tried: '/opt/homebrew/lib/librwkv.dylib' (no such file), '/Users/joakimeriksson/.wasmedge/lib/librwkv.dylib' (no such file), '/Users/joakimeriksson/work/rwkv.cpp/librwkv.dylib' (mach-o file, but is an incompatible architecture (have 'arm64', need 'x86_64')), '/System/Volumes/Preboot/Cryptexes/OS/Users/joakimeriksson/work/rwkv.cpp/librwkv.dylib' (no such file), '/Users/joakimeriksson/work/rwkv.cpp/librwkv.dylib' (mach-o file, but is an incompatible architecture (have 'arm64', need 'x86_64'))

I have Mac with non-intel CPU. Somehow the build-scripts detects ARM64 but when running assumes x86.

[joakimeriksson]

My problem was that my conda environment used an x86 python while the compiler did compile for arm64. Replacing the conda environment with an arm64 worked.