export error v2.3.1

[chrizzlemadizzle]

Hey! I'm receiving an error when trying to export a model with v2.3.1 using the following configs:

!/content/miniconda/bin/rave preprocess --input_path $dataset --output_path $preprocessed_dataset --channels 2 !/content/miniconda/bin/rave train --config v3 --config wasserstein --db_path $preprocessed_dataset --name $name --val_every 500 --channels 2

!/content/miniconda/bin/rave export --run $model_dir --streaming --channels 2 --fidelity 0.999

the error:

INFO:root:library loading INFO:root:DEBUG I1222 19:43:17.716103 140450175820864 export.py:495] building rave /content/miniconda/lib/python3.9/site-packages/torch/nn/utils/weight_norm.py:30: UserWarning: torch.nn.utils.weight_norm is deprecated in favor of torch.nn.utils.parametrizations.weight_norm. warnings.warn("torch.nn.utils.weight_norm is deprecated in favor of torch.nn.utils.parametrizations.weight_norm.") /content/miniconda/lib/python3.9/site-packages/torchaudio/transforms/_transforms.py:94: UserWarning: return_complex argument is now deprecated and is not effective.torchaudio.transforms.Spectrogram(power=None) always returns a tensor with complex dtype. Please remove the argument in the function call. warnings.warn( I1222 19:43:19.216365 140450175820864 export.py:505] model found : /content/drive/MyDrive/AI/RAVE/vivaZweiTraining/2023-12-22-testing2-3-1/runs/viva_35fc26584b/version_0/checkpoints/epoch-epoch=0020.ckpt Traceback (most recent call last): File "/content/miniconda/bin/rave", line 8, in <module> sys.exit(main()) File "/content/miniconda/lib/python3.9/site-packages/scripts/main_cli.py", line 38, in main app.run(export.main) File "/content/miniconda/lib/python3.9/site-packages/absl/app.py", line 308, in run _run_main(main, args) File "/content/miniconda/lib/python3.9/site-packages/absl/app.py", line 254, in _run_main sys.exit(main(argv)) File "/content/miniconda/lib/python3.9/site-packages/scripts/export.py", line 513, in main pretrained.load_state_dict( File "/content/miniconda/lib/python3.9/site-packages/torch/nn/modules/module.py", line 2152, in load_state_dict raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format( RuntimeError: Error(s) in loading state_dict for RAVE: size mismatch for encoder.encoder.net.0.weight_v: copying a param with shape torch.Size([96, 32, 7]) from checkpoint, the shape in current model is torch.Size([96, 16, 7]). size mismatch for decoder.net.32.weight_g: copying a param with shape torch.Size([64, 1, 1]) from checkpoint, the shape in current model is torch.Size([32, 1, 1]). size mismatch for decoder.net.32.weight_v: copying a param with shape torch.Size([64, 96, 7]) from checkpoint, the shape in current model is torch.Size([32, 96, 7]). size mismatch for discriminator.discriminators.0.convs.0.0.weight_v: copying a param with shape torch.Size([32, 2, 5, 1]) from checkpoint, the shape in current model is torch.Size([32, 1, 5, 1]). size mismatch for discriminator.discriminators.1.convs.0.0.weight_v: copying a param with shape torch.Size([32, 2, 5, 1]) from checkpoint, the shape in current model is torch.Size([32, 1, 5, 1]). size mismatch for discriminator.discriminators.2.convs.0.0.weight_v: copying a param with shape torch.Size([32, 2, 5, 1]) from checkpoint, the shape in current model is torch.Size([32, 1, 5, 1]). size mismatch for discriminator.discriminators.3.convs.0.0.weight_v: copying a param with shape torch.Size([32, 2, 5, 1]) from checkpoint, the shape in current model is torch.Size([32, 1, 5, 1]). size mismatch for discriminator.discriminators.4.convs.0.0.weight_v: copying a param with shape torch.Size([32, 2, 5, 1]) from checkpoint, the shape in current model is torch.Size([32, 1, 5, 1]). size mismatch for discriminator.discriminators.5.band_convs.0.0.0.weight_v: copying a param with shape torch.Size([32, 4, 3, 9]) from checkpoint, the shape in current model is torch.Size([32, 2, 3, 9]). size mismatch for discriminator.discriminators.5.band_convs.1.0.0.weight_v: copying a param with shape torch.Size([32, 4, 3, 9]) from checkpoint, the shape in current model is torch.Size([32, 2, 3, 9]). size mismatch for discriminator.discriminators.5.band_convs.2.0.0.weight_v: copying a param with shape torch.Size([32, 4, 3, 9]) from checkpoint, the shape in current model is torch.Size([32, 2, 3, 9]). size mismatch for discriminator.discriminators.5.band_convs.3.0.0.weight_v: copying a param with shape torch.Size([32, 4, 3, 9]) from checkpoint, the shape in current model is torch.Size([32, 2, 3, 9]). size mismatch for discriminator.discriminators.5.band_convs.4.0.0.weight_v: copying a param with shape torch.Size([32, 4, 3, 9]) from checkpoint, the shape in current model is torch.Size([32, 2, 3, 9]). size mismatch for discriminator.discriminators.6.band_convs.0.0.0.weight_v: copying a param with shape torch.Size([32, 4, 3, 9]) from checkpoint, the shape in current model is torch.Size([32, 2, 3, 9]). size mismatch for discriminator.discriminators.6.band_convs.1.0.0.weight_v: copying a param with shape torch.Size([32, 4, 3, 9]) from checkpoint, the shape in current model is torch.Size([32, 2, 3, 9]). size mismatch for discriminator.discriminators.6.band_convs.2.0.0.weight_v: copying a param with shape torch.Size([32, 4, 3, 9]) from checkpoint, the shape in current model is torch.Size([32, 2, 3, 9]). size mismatch for discriminator.discriminators.6.band_convs.3.0.0.weight_v: copying a param with shape torch.Size([32, 4, 3, 9]) from checkpoint, the shape in current model is torch.Size([32, 2, 3, 9]). size mismatch for discriminator.discriminators.6.band_convs.4.0.0.weight_v: copying a param with shape torch.Size([32, 4, 3, 9]) from checkpoint, the shape in current model is torch.Size([32, 2, 3, 9]). size mismatch for discriminator.discriminators.7.band_convs.0.0.0.weight_v: copying a param with shape torch.Size([32, 4, 3, 9]) from checkpoint, the shape in current model is torch.Size([32, 2, 3, 9]). size mismatch for discriminator.discriminators.7.band_convs.1.0.0.weight_v: copying a param with shape torch.Size([32, 4, 3, 9]) from checkpoint, the shape in current model is torch.Size([32, 2, 3, 9]). size mismatch for discriminator.discriminators.7.band_convs.2.0.0.weight_v: copying a param with shape torch.Size([32, 4, 3, 9]) from checkpoint, the shape in current model is torch.Size([32, 2, 3, 9]). size mismatch for discriminator.discriminators.7.band_convs.3.0.0.weight_v: copying a param with shape torch.Size([32, 4, 3, 9]) from checkpoint, the shape in current model is torch.Size([32, 2, 3, 9]). size mismatch for discriminator.discriminators.7.band_convs.4.0.0.weight_v: copying a param with shape torch.Size([32, 4, 3, 9]) from checkpoint, the shape in current model is torch.Size([32, 2, 3, 9]).

[augustross3]

Having this same error when exporting on v2.3

[spwilson2]

Yeah, I'm seeing issues exporting any models w/ two channels. Single channel seems to work fine.

[spwilson2]

In mycase it looks like the following wasn't stored in the generated .gin files for some reason:

model.RAVE.n_channels = 2

When I added that to the .gin file half of the reported errors went away and I was left w/

Traceback (most recent call last):
  File "C:\Users\spwil\AppData\Local\Programs\Python\Python310\Scripts\rave-script.py", line 33, in <module>
    sys.exit(load_entry_point('acids-rave==2.3.1', 'console_scripts', 'rave')())
  File "C:\Users\spwil\AppData\Local\Programs\Python\Python310\lib\site-packages\scripts\main_cli.py", line 38, in main
    app.run(export.main)
  File "C:\Users\spwil\AppData\Local\Programs\Python\Python310\lib\site-packages\absl\app.py", line 308, in run
    _run_main(main, args)
  File "C:\Users\spwil\AppData\Local\Programs\Python\Python310\lib\site-packages\absl\app.py", line 254, in _run_main
    sys.exit(main(argv))
  File "C:\Users\spwil\AppData\Local\Programs\Python\Python310\lib\site-packages\scripts\export.py", line 514, in main
    pretrained.load_state_dict(
  File "C:\Users\spwil\AppData\Local\Programs\Python\Python310\lib\site-packages\torch\nn\modules\module.py", line 1671, in load_state_dict
    raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for RAVE:
        size mismatch for latent_pca: copying a param with shape torch.Size([16, 16]) from checkpoint, the shape in current model is torch.Size([2, 2]).
        size mismatch for latent_mean: copying a param with shape torch.Size([16]) from checkpoint, the shape in current model is torch.Size([2]).
        size mismatch for fidelity: copying a param with shape torch.Size([16]) from checkpoint, the shape in current model is torch.Size([2]).

That I was able to workaround with that with the following patch on top of v2.3.1:

diff --git a/rave/model.py b/rave/model.py
index d30002b..38b06fc 100644
--- a/rave/model.py
+++ b/rave/model.py
@@ -192,7 +192,7 @@ class RAVE(pl.LightningModule):
         self.multiband_audio_distance = multiband_audio_distance()

         self.gan_loss = gan_loss
-
+        latent_size = 16
         self.register_buffer("latent_pca", torch.eye(latent_size))
         self.register_buffer("latent_mean", torch.zeros(latent_size))
         self.register_buffer("fidelity", torch.zeros(latent_size))

Edit:
It does seem to work with the M4L nn~ object, so this at least works as a workaround for now...

[maksimay]

Same for me thanks so much for the workaround.

I also managed to export my 3channel model with this fix

adding model.RAVE.n_channels = 3 in the config.gin
and by changing the lines of the export.py accordingly:

self.register_method(
          "encode",
          in_channels=3, #my number of channels
          in_ratio=1,
          out_channels=3, #my number of channels
          out_ratio=ratio_encode,
          input_labels=['(signal) Input audio signal'],
          output_labels=[
              f'(signal) Latent dimension {i}'
              for i in range(self.latent_size)
          ],
      )
      self.register_method(
          "decode",
          in_channels=3, #my number of channels
          in_ratio=ratio_encode,
          out_channels=3, #my number of channels
          out_ratio=1,
          input_labels=[
              f'(signal) Latent dimension {i}'
              for i in range(self.latent_size)
          ],
          output_labels=[
              f'(signal) Reconstructed audio signal {channel}'
              for channel in channels
          ],
      )

      self.register_method(
          "forward",
          in_channels=3, #my number of channels
          in_ratio=1,
          out_channels=3, #my number of channels
          out_ratio=1,
          input_labels=['(signal) Input audio signal'],
          output_labels=[
              f'(signal) Reconstructed audio signal {channel}'
              for channel in channels
          ],
      )