search

ggerganov / whisper.cpp

Port of OpenAI's Whisper model in C/C++
MIT License
35.42k stars 3.61k forks source link

Error in convert-whisper-to-coreml.py ? #1009

Open jaybinks opened 1 year ago

jaybinks commented 1 year ago

On a nearly brand-new mac mini M2, I get the following error trying to generate the coreml files.

Any ideas / suggestions would be appreciated.


/opt/homebrew/lib/python3.11/site-packages/whisper/timing.py:57: NumbaDeprecationWarning: The 'nopython' keyword argument was not supplied to the 'numba.jit' decorator. The implicit default value for this argument is currently False, but it will be changed to True in Numba 0.59.0. See https://numba.readthedocs.io/en/stable/reference/deprecation.html#deprecation-of-object-mode-fall-back-behaviour-when-using-jit for details.
  @numba.jit

ModelDimensions(n_mels=80, n_audio_ctx=1500, n_audio_state=1280, n_audio_head=20, n_audio_layer=32, n_vocab=51865, n_text_ctx=448, n_text_state=1280, n_text_head=20, n_text_layer=32)
/opt/homebrew/lib/python3.11/site-packages/whisper/model.py:166: TracerWarning: Converting a tensor to a Python boolean might cause the trace to be incorrect. We cant record the data flow of Python values, so this value will be treated as a constant in the future. This means that the trace might not generalize to other inputs!
  assert x.shape[1:] == self.positional_embedding.shape, "incorrect audio shape"

Converting PyTorch Frontend ==> MIL Ops: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████▉| 2613/2614 [00:00<00:00, 3476.92 ops/s]
Running MIL frontend_pytorch pipeline: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 5/5 [00:00<00:00, 33.49 passes/s]
Running MIL default pipeline: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 57/57 [00:12<00:00,  4.50 passes/s]
Running MIL backend_mlprogram pipeline: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 10/10 [00:00<00:00, 368.28 passes/s]

Traceback (most recent call last):
  File "/Users/jaybinks/src/whisper.cpp/models/convert-whisper-to-coreml.py", line 323, in <module>
    encoder = convert_encoder(hparams, encoder, quantize=args.quantize)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/jaybinks/src/whisper.cpp/models/convert-whisper-to-coreml.py", line 259, in convert_encoder
    model = ct.convert(
            ^^^^^^^^^^^
  File "/opt/homebrew/lib/python3.11/site-packages/coremltools/converters/_converters_entry.py", line 492, in convert
    mlmodel = mil_convert(
              ^^^^^^^^^^^^
  File "/opt/homebrew/lib/python3.11/site-packages/coremltools/converters/mil/converter.py", line 188, in mil_convert
    return _mil_convert(model, convert_from, convert_to, ConverterRegistry, MLModel, compute_units, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/homebrew/lib/python3.11/site-packages/coremltools/converters/mil/converter.py", line 212, in _mil_convert
    proto, mil_program = mil_convert_to_proto(
                         ^^^^^^^^^^^^^^^^^^^^^
  File "/opt/homebrew/lib/python3.11/site-packages/coremltools/converters/mil/converter.py", line 303, in mil_convert_to_proto
    out = backend_converter(prog, **kwargs)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/homebrew/lib/python3.11/site-packages/coremltools/converters/mil/converter.py", line 130, in __call__
    return backend_load(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/homebrew/lib/python3.11/site-packages/coremltools/converters/mil/backend/mil/load.py", line 283, in load
    raise RuntimeError("BlobWriter not loaded")

RuntimeError: BlobWriter not loaded

coremlc: error: Model does not exist at models/coreml-encoder-large.mlpackage -- file:///Users/jaybinks/src/whisper.cpp/

mv: rename models/coreml-encoder-large.mlmodelc to models/ggml-large-encoder.mlmodelc: No such file or directory
jaybinks commented 1 year ago

The same issue happens with medium and small models.

jaybinks commented 1 year ago

Incase it helps 

pip list
Package            Version
------------------ ---------
ane-transformers   0.1.1
certifi            2023.5.7
charset-normalizer 3.1.0
coremltools        6.3.0
ffmpeg-python      0.2.0
filelock           3.12.1
fsspec             2023.6.0
future             0.18.3
huggingface-hub    0.15.1
idna               3.4
Jinja2             3.1.2
llvmlite           0.40.1rc1
MarkupSafe         2.1.3
more-itertools     9.1.0
mpmath             1.3.0
networkx           3.1
numba              0.57.0
numpy              1.24.3
openai-whisper     20230314
packaging          23.1
pip                23.1.2
protobuf           3.20.1
PyYAML             6.0
regex              2023.6.3
requests           2.31.0
safetensors        0.3.1
setuptools         67.6.1
sympy              1.12
tiktoken           0.3.1
tokenizers         0.13.3
torch              2.0.0
tqdm               4.65.0
transformers       4.30.1
typing_extensions  4.6.3
urllib3            2.0.3
wheel              0.40.0
brew list --versions

aom 3.6.1
brotli 1.0.9
ca-certificates 2023-05-30
cairo 1.16.0_5
chromaprint 1.5.1_1
cjson 1.7.15
coreutils 9.3
dav1d 1.2.1
ffmpeg@4 4.4.4
fftw 3.3.10_1
flac 1.4.2
fontconfig 2.14.2
freetype 2.13.0_1
frei0r 1.8.0
fribidi 1.0.13
gcc 13.1.0
gettext 0.21.1
giflib 5.2.1
git 2.41.0
glib 2.76.3
gmp 6.2.1_1
gnutls 3.8.0
graphite2 1.3.14
harfbuzz 7.3.0
highway 1.0.4
htop 3.2.2
hwloc 2.9.1
icu4c 72.1
imath 3.1.9
isl 0.26
jpeg-turbo 2.1.5.1
jpeg-xl 0.8.1_3
lame 3.100
leptonica 1.82.0_2
libarchive 3.6.2_1
libass 0.17.1
libb2 0.98.1
libbluray 1.3.4
libevent 2.1.12
libidn2 2.3.4_1
libmpc 1.3.1
libnghttp2 1.54.0
libogg 1.3.5
libpng 1.6.39
librist 0.2.7_3
libsamplerate 0.2.2
libsndfile 1.2.0_1
libsodium 1.0.18_1
libsoxr 0.1.3
libtasn1 4.19.0
libtiff 4.5.0
libunibreak 5.1
libunistring 1.1
libvidstab 1.1.1
libvmaf 2.3.1
libvorbis 1.3.7
libvpx 1.13.0
libx11 1.8.4
libxau 1.0.11
libxcb 1.15_1
libxdmcp 1.1.4
libxext 1.3.5
libxrender 0.9.11
little-cms2 2.15
lz4 1.9.4
lzo 2.10
mad 0.15.1b
mbedtls 3.4.0
mpdecimal 2.5.1
mpfr 4.2.0-p9
mpg123 1.31.3
ncurses 6.4
nettle 3.9.1
open-mpi 4.1.5
opencore-amr 0.1.6
openexr 3.1.8_1
openjpeg 2.5.0_1
openssl@1.1 1.1.1u
openssl@3 3.1.1
opus 1.4
opusfile 0.12
p11-kit 0.24.1_1
pango 1.50.14
pcre2 10.42
pixman 0.42.2
pkg-config 0.29.2_3
python@3.11 3.11.4
rav1e 0.6.6
readline 8.2.1
rubberband 3.2.1
sdl2 2.26.5
snappy 1.1.10
sox 14.4.2_5
speex 1.2.1
sqlite 3.42.0
srt 1.5.1
tesseract 5.3.1
theora 1.1.1
unbound 1.17.1
webp 1.3.0_1
wget 1.21.4
x264 r3095
x265 3.5
xorgproto 2022.2
xvid 1.3.7
xz 5.4.3
zeromq 4.3.4
zimg 3.0.4
zstd 1.5.5
miniconda py310_23.3.1-0
jaybinks commented 1 year ago

More info about my env : 

jaybinks@Jays-Mac-mini whisper.cpp % conda info

     active environment : None
       user config file : /Users/jaybinks/.condarc
 populated config files :
          conda version : 23.3.1
    conda-build version : not installed
         python version : 3.10.10.final.0
       virtual packages : __archspec=1=arm64
                          __osx=13.2.1=0
                          __unix=0=0
       base environment : /opt/homebrew/Caskroom/miniconda/base  (writable)
      conda av data dir : /opt/homebrew/Caskroom/miniconda/base/etc/conda
  conda av metadata url : None
           channel URLs : https://repo.anaconda.com/pkgs/main/osx-arm64
                          https://repo.anaconda.com/pkgs/main/noarch
                          https://repo.anaconda.com/pkgs/r/osx-arm64
                          https://repo.anaconda.com/pkgs/r/noarch
          package cache : /opt/homebrew/Caskroom/miniconda/base/pkgs
                          /Users/jaybinks/.conda/pkgs
       envs directories : /opt/homebrew/Caskroom/miniconda/base/envs
                          /Users/jaybinks/.conda/envs
               platform : osx-arm64
             user-agent : conda/23.3.1 requests/2.28.1 CPython/3.10.10 Darwin/22.3.0 OSX/13.2.1
                UID:GID : 501:20
             netrc file : None
           offline mode : False

jaybinks@Jays-Mac-mini whisper.cpp % pip freeze
ane-transformers==0.1.1
certifi==2023.5.7
charset-normalizer==3.1.0
coremltools==6.3.0
ffmpeg-python==0.2.0
filelock==3.12.1
fsspec==2023.6.0
future==0.18.3
huggingface-hub==0.15.1
idna==3.4
Jinja2==3.1.2
llvmlite==0.40.1rc1
MarkupSafe==2.1.3
more-itertools==9.1.0
mpmath==1.3.0
networkx==3.1
numba==0.57.0
numpy==1.24.3
openai-whisper==20230314
packaging==23.1
protobuf==3.20.1
PyYAML==6.0
regex==2023.6.3
requests==2.31.0
safetensors==0.3.1
sympy==1.12
tiktoken==0.3.1
tokenizers==0.13.3
torch==2.0.0
tqdm==4.65.0
transformers==4.30.1
typing_extensions==4.6.3
urllib3==2.0.3
jaybinks@Jays-Mac-mini whisper.cpp %
jaybinks commented 1 year ago

I managed to get around this by downloading the coreml models from the URL here : https://github.com/ggerganov/whisper.cpp/pull/566

however it still might be good to know if this is a me problem, or if the instructions need to be refined some more.

andrei-zgirvaci commented 1 year ago

@jaybinks I believe the error comes from using Python 3.11 which currently is not compatible with the coremltools master branch. https://github.com/apple/coremltools/issues/1730#issuecomment-1382615962

The solution is to either update to Coremltools 7.0b1 or downgrade to Python 3.10. I tried Coremltools 7.0b1 and it converted successfully.

ChloeLo27 commented 1 year ago

I have to second that I ran into a similar problem, and I was on Python 3.10.11, followed the setup instruction to a T. The problem occurred when I tried to build the CoreML version of medium.en:

whisper.cpp % ./models/generate-coreml-model.sh medium.en
Torch version 2.0.1 has not been tested with coremltools. You may run into unexpected errors. Torch 2.0.0 is the most recent version that has been tested.
Traceback (most recent call last):
  File "<snip>/whisper.cpp/models/convert-whisper-to-coreml.py", line 10, in <module>
    from ane_transformers.reference.layer_norm import LayerNormANE as LayerNormANEBase
ModuleNotFoundError: No module named 'ane_transformers.reference'
coremlc: error: Model does not exist at models/coreml-encoder-medium.en.mlpackage -- file:///<snip>/whisper.cpp/
mv: rename models/coreml-encoder-medium.en.mlmodelc to models/ggml-medium.en-encoder.mlmodelc: No such file or directory

Any insight would be appreciated.

I tried to find the CoreML model download link in https://github.com/ggerganov/whisper.cpp/pull/566 as suggested by @jaybinks, but I seem to get lost in that long PR thread so I would appreciate a more precise pointer to where I can find it...

jaybinks commented 1 year ago

I tried to find the CoreML model download link in #566 as suggested by @jaybinks, but I seem to get lost in that long PR thread so I would appreciate a more precise pointer to where I can find it...

https://drive.proton.me/urls/66XZ3PRXP4#DkN12NuE2mpY

patriotyk commented 1 year ago

I had the same error when using python 3.11. After using venv with python 3.10 issue has gone.

AbeEstrada commented 1 year ago

This works for me using Python 3.11

pip uninstall coremltools
pip install coremltools==7.0b2
./models/generate-coreml-model.sh base.en
gkovacsp commented 1 year ago

Based on the above discussion this should absolutely work - but it does not. This is my conda environment on an Apple M1 Pro:

(python310) kpg-mcb:whisper.cpp kpg$ conda list
# packages in environment at /Users/kpg/miniconda3/envs/python310:
#
# Name                    Version                   Build  Channel
ane-transformers          0.1.3                    pypi_0    pypi
attrs                     23.1.0                   pypi_0    pypi
bzip2                     1.0.8                h620ffc9_4  
ca-certificates           2023.08.22           hca03da5_0  
cattrs                    23.1.2                   pypi_0    pypi
certifi                   2023.7.22                pypi_0    pypi
charset-normalizer        3.3.0                    pypi_0    pypi
coremltools               7.0                      pypi_0    pypi
exceptiongroup            1.1.3                    pypi_0    pypi
filelock                  3.12.4                   pypi_0    pypi
fsspec                    2023.9.2                 pypi_0    pypi
huggingface-hub           0.17.3                   pypi_0    pypi
idna                      3.4                      pypi_0    pypi
libffi                    3.4.4                hca03da5_0  
llvmlite                  0.41.0                   pypi_0    pypi
more-itertools            10.1.0                   pypi_0    pypi
mpmath                    1.3.0                    pypi_0    pypi
ncurses                   6.4                  h313beb8_0  
numba                     0.58.0                   pypi_0    pypi
numpy                     1.25.2                   pypi_0    pypi
openai-whisper            20230918                 pypi_0    pypi
openssl                   1.1.1w               h1a28f6b_0  
packaging                 23.2                     pypi_0    pypi
pip                       23.2.1          py310hca03da5_0  
protobuf                  3.20.1                   pypi_0    pypi
pyaml                     23.9.7                   pypi_0    pypi
python                    3.10.0               hbdb9e5c_5  
pyyaml                    6.0.1                    pypi_0    pypi
readline                  8.2                  h1a28f6b_0  
regex                     2023.10.3                pypi_0    pypi
requests                  2.31.0                   pypi_0    pypi
safetensors               0.4.0                    pypi_0    pypi
setuptools                68.0.0          py310hca03da5_0  
sqlite                    3.41.2               h80987f9_0  
sympy                     1.12                     pypi_0    pypi
tiktoken                  0.3.3                    pypi_0    pypi
tk                        8.6.12               hb8d0fd4_0  
tokenizers                0.14.1                   pypi_0    pypi
torch                     1.11.0                   pypi_0    pypi
tqdm                      4.66.1                   pypi_0    pypi
transformers              4.34.0                   pypi_0    pypi
typing-extensions         4.8.0                    pypi_0    pypi
tzdata                    2023c                h04d1e81_0  
urllib3                   2.0.6                    pypi_0    pypi
wheel                     0.41.2          py310hca03da5_0  
xz                        5.4.2                h80987f9_0  
zlib                      1.2.13               h5a0b063_0

Error:

Converting PyTorch Frontend ==> MIL Ops: 100%|███████████████████████████████████████████████████████████████████████████████▊| 533/534 [00:00<00:00, 5391.68 ops/s]
Running MIL frontend_pytorch pipeline: 100%|████████████████████████████████████████████████████████████████████████████████████| 5/5 [00:00<00:00, 576.57 passes/s]
Running MIL default pipeline: 100%|████████████████████████████████████████████████████████████████████████████████████████████| 57/57 [00:00<00:00, 81.39 passes/s]
Running MIL backend_mlprogram pipeline: 100%|████████████████████████████████████████████████████████████████████████████████| 10/10 [00:00<00:00, 2119.62 passes/s]
Traceback (most recent call last):
  File "/Users/kpg/python/whisper.cpp/models/convert-whisper-to-coreml.py", line 323, in <module>
    encoder = convert_encoder(hparams, encoder, quantize=args.quantize)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/kpg/python/whisper.cpp/models/convert-whisper-to-coreml.py", line 259, in convert_encoder
    model = ct.convert(
            ^^^^^^^^^^^
  File "/opt/homebrew/lib/python3.11/site-packages/coremltools/converters/_converters_entry.py", line 492, in convert
    mlmodel = mil_convert(
              ^^^^^^^^^^^^
  File "/opt/homebrew/lib/python3.11/site-packages/coremltools/converters/mil/converter.py", line 188, in mil_convert
    return _mil_convert(model, convert_from, convert_to, ConverterRegistry, MLModel, compute_units, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/homebrew/lib/python3.11/site-packages/coremltools/converters/mil/converter.py", line 212, in _mil_convert
    proto, mil_program = mil_convert_to_proto(
                         ^^^^^^^^^^^^^^^^^^^^^
  File "/opt/homebrew/lib/python3.11/site-packages/coremltools/converters/mil/converter.py", line 303, in mil_convert_to_proto
    out = backend_converter(prog, **kwargs)
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/homebrew/lib/python3.11/site-packages/coremltools/converters/mil/converter.py", line 130, in __call__
    return backend_load(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/homebrew/lib/python3.11/site-packages/coremltools/converters/mil/backend/mil/load.py", line 283, in load
    raise RuntimeError("BlobWriter not loaded")
RuntimeError: BlobWriter not loaded
coremlc: error: Model does not exist at models/coreml-encoder-base.en.mlpackage -- file:///Users/kpg/python/whisper.cpp/
mv: rename models/coreml-encoder-base.en.mlmodelc to models/ggml-base.en-encoder.mlmodelc: No such file or directory

Tne direct download to coreml models does not work any more, is there another location to download them?

hanfengv commented 2 months ago

pip install numpy==1.26.4 it's helpful for me