using `convert-h5-to-ggml.py` I get this error : FileNotFoundError: [Errno 2] No such file or directory: 'whisper.cpp/whisper/assets/mel_filters.npz'

[raphael10-collab]

I want to convert this pythorch model : https://huggingface.co/whispy/whisper_italian into ggml model

But when using convert-h5-to-ggml.py I get this error :

(.venv) riccardo@riccardo-HP-Laptop-15-da0xxx:~/Downloads$ python3 ./whisper.cpp/models/convert-h5-to-ggml.py ./whisper_italian/ ./whisper.cpp/ .
Traceback (most recent call last):
  File "/home/riccardo/Downloads/./whisper.cpp/models/convert-h5-to-ggml.py", line 96, in <module>
    with np.load(os.path.join(dir_whisper, "whisper/assets", "mel_filters.npz")) as f:
         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/riccardo/Downloads/whisper.cpp/models/.venv/lib/python3.12/site-packages/numpy/lib/_npyio_impl.py", line 451, in load
    fid = stack.enter_context(open(os.fspath(file), "rb"))
                              ^^^^^^^^^^^^^^^^^^^^^^^^^^^
FileNotFoundError: [Errno 2] No such file or directory: 'whisper.cpp/whisper/assets/mel_filters.npz'

I tried also with the ./models/README example but the same error:

(.venv) riccardo@riccardo-HP-Laptop-15-da0xxx:~/Downloads$ git clone https://huggingface.co/openai/whisper-medium
Cloning into 'whisper-medium'...
remote: Enumerating objects: 175, done.
remote: Counting objects: 100% (6/6), done.
remote: Compressing objects: 100% (6/6), done.
remote: Total 175 (delta 1), reused 0 (delta 0), pack-reused 169 (from 1)
Receiving objects: 100% (175/175), 2.88 MiB | 6.30 MiB/s, done.
Resolving deltas: 100% (102/102), done.
Filtering content: 100% (4/4), 11.38 GiB | 80.20 MiB/s, done.



(.venv) riccardo@riccardo-HP-Laptop-15-da0xxx:~/Downloads$ python3 ./whisper.cpp/models/convert-h5-to-ggml.py ./whisper-medium/ ./whisper .
Traceback (most recent call last):
  File "/home/riccardo/Downloads/./whisper.cpp/models/convert-h5-to-ggml.py", line 96, in <module>
    with np.load(os.path.join(dir_whisper, "whisper/assets", "mel_filters.npz")) as f:
         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/riccardo/Downloads/whisper.cpp/models/.venv/lib/python3.12/site-packages/numpy/lib/_npyio_impl.py", line 451, in load
    fid = stack.enter_context(open(os.fspath(file), "rb"))
                              ^^^^^^^^^^^^^^^^^^^^^^^^^^^
FileNotFoundError: [Errno 2] No such file or directory: 'whisper/whisper/assets/mel_filters.npz'

How to make it work?

[foldl]

Usage: convert-h5-to-ggml.py dir_model path-to-whisper-repo dir-output [use-f32]

You need to pass the path to whisper repo, but not whisper.cpp as the second parameter.

You can find mel_filters.npz there:

https://github.com/openai/whisper/blob/main/whisper/assets/

[markuman]

python convert-h5-to-ggml.py ../../tiny-ger/ ../../whisper/ ../../out/
Traceback (most recent call last):
  File "/home/m/git/lekker/markus.bergholz/tmp/gguf_convert/whisper.cpp/models/convert-h5-to-ggml.py", line 119, in <module>
    fout.write(struct.pack("i", hparams["max_length"]))
               ~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^
struct.error: required argument is not an integer

Basically I try to conert this https://huggingface.co/primeline/whisper-tiny-german-1224 into a single .bin file.
Any ideas @foldl?

[markuman]

Ok, this PR solves my issue: #2840