spectraldoy / music-transformer Goto Github PK

Heya, I faced this error while trying to generate some MIDIs based on your pretrained weights. Do you know how I can fix this? I feel that it's something to do with the PyTorch version but I have installed the versions as listed in requirements.txt.

PS C:\Users\USER\Desktop\music-transformer-main> python generate.py models/chopintransformerv5.pt .../gen_audio.mid -v
Greedy decoding...
Traceback (most recent call last):
File "C:\Users\USER\Desktop\music-transformer-main\generate.py", line 282, in
generate(model_=music_transformer, inp=[""], save_path=args.save_path,
File "C:\Users\USER\Desktop\music-transformer-main\generate.py", line 229, in generate
token_ids = greedy_decode(model=model_, inp=inp, mode=mode, temperature=temperature, k=k)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\USER\Desktop\music-transformer-main\generate.py", line 109, in greedy_decode
predictions = model(inp, mask=create_mask(inp, n))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\USER\AppData\Local\Programs\Python\Python311\Lib\site-packages\torch\nn\modules\module.py", line 1511, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\USER\AppData\Local\Programs\Python\Python311\Lib\site-packages\torch\nn\modules\module.py", line 1520, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\USER\Desktop\music-transformer-main\model.py", line 107, in forward
x = self.decoder(x, memory=None, tgt_mask=mask)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\USER\AppData\Local\Programs\Python\Python311\Lib\site-packages\torch\nn\modules\module.py", line 1511, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\USER\AppData\Local\Programs\Python\Python311\Lib\site-packages\torch\nn\modules\module.py", line 1520, in _call_impl
return forward_call(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\USER\AppData\Local\Programs\Python\Python311\Lib\site-packages\torch\nn\modules\transformer.py", line 461, in forward
seq_len = _get_seq_len(tgt, self.layers[0].self_attn.batch_first)
^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\USER\AppData\Local\Programs\Python\Python311\Lib\site-packages\torch\nn\modules\module.py", line 1688, in getattr
raise AttributeError(f"'{type(self).name}' object has no attribute '{name}'")
AttributeError: 'DecoderLayer' object has no attribute 'self_attn'

Given appropriately normalised MIDI files, this transformer produces results that are truly astonishing, a head and shoulders above all the other similar projects on github.
Samples of generated Baroque music attached.
samples.zip

First of all, this is a great repo and it was super helpful in understanding the music transformer! I had some questions about training. For the pretrained models you included, how many epochs did you train for, and approximately how long did training take? Do you have any graphs from these runs that you could share, and what training / validation loss did you achieve?