vasista22

/

whisper-tamil-small

like 4

Hello,
I'm trying to convert the model into ggml format to use in whisper.cpp. Unfortunately when I run the command as recommended over at whisper.cpp:

!python3 ggml-to-pt.py pytorch_model.bin whisper.cpp/ ggml-tamil-small-vasista.bin

I get the following error:

Magic number: 67324752
Vocab size: 134742016
Audio context size: 0
Audio state size: 0
Audio head size: 0
Audio layer size: 0
Text context size: 1048576
Text head size: 1986619491
Mel size: 1882087796
Filters shape 0: 1515847694
Filters shape 1: 1515870810
Traceback (most recent call last):
  File "/content/ggml-to-pt.py", line 48, in <module>
    mel_filters = np.zeros((filters_shape_0, filters_shape_1))
ValueError: array is too big; `arr.size * arr.dtype.itemsize` is larger than the maximum possible size.

Any recommendations on how to fix this? I'm having the same issue with the medium model as well.