gigant

/

whisper-medium-romanian

like 14

i've never used this, i'll take a look at it :)

I also tried to compile into ggml bin using something like this:
python models/convert-pt-to-ggml.py ~/AI-test/voice/whisper-medium-romanian/checkpoint-1000/optimizer.pt ../whisper ./models/whisper-medium
but got the infamous error:

    hparams = checkpoint["dims"]
KeyError: 'dims' ```

on a "normal" model after *torch.load* it returns  an array similar to this :

{'dims': {'n_mels': 80, 'n_vocab': 51864, 'n_audio_ctx': 1500, 'n_audio_state': 384, 'n_audio_head': 6, 'n_audio_layer': 4, 'n_text_ctx': 448, 'n_text_state': 384, 'n_text_head': 6, 'n_text_layer': 4}, 'model_state_dict':

Later edit
I was using the wrong script . It was written in the documentation: 

python models/convert-h5-to-ggml.py /mypool/par3disk/mihaim/AI/voice/whisper-medium-romanian/checkpoint-3500 ../whisper ./models

So it's actually working . 

nice @mmol . I didn't have time to check all of that but it seems that you made it work :)
if you want to share the ggml binary you created do not hesitate to open a PR on this model

I also tried to compile into ggml bin using something like this:
python models/convert-pt-to-ggml.py ~/AI-test/voice/whisper-medium-romanian/checkpoint-1000/optimizer.pt ../whisper ./models/whisper-medium
but got the infamous error:

    hparams = checkpoint["dims"]
KeyError: 'dims' ```

on a "normal" model after *torch.load* it returns  an array similar to this :

{'dims': {'n_mels': 80, 'n_vocab': 51864, 'n_audio_ctx': 1500, 'n_audio_state': 384, 'n_audio_head': 6, 'n_audio_layer': 4, 'n_text_ctx': 448, 'n_text_state': 384, 'n_text_head': 6, 'n_text_layer': 4}, 'model_state_dict':

Later edit
I was using the wrong script . It was written in the documentation: 

python models/convert-h5-to-ggml.py /mypool/par3disk/mihaim/AI/voice/whisper-medium-romanian/checkpoint-3500 ../whisper ./models

So it's actually working .

Ai reusit sa-l faci sa mearga in whisper?