'SemanticTransformerWrapper' object has no attribute 'embed_text'

[jinyuli]

I'm trying to generate sound with text, but got this:

AttributeError: 'SemanticTransformerWrapper' object has no attribute 'embed_text'

and I didn't find the method embed_text in class SemanticTransformerWrapper either.

[lucidrains]

@jinyuli oh nice, you must have gotten to the final stage? could you try 1.1.4?

[jinyuli]

thanks, it's fixed.

but I got another issue, while generating audio with text(running on google Colab):

• if cuda is used, then got error like "GET was unable to find an engine to execute" when torchaudio trying to save the audio file, after googling, the cause may be CUDA version incompatible, Colab cuda version is 12.0, Pytorch's is 11.8, I'm not sure whether it's a problem.
• if cpu is used, then got error said that all tensors should use the same device, but I've set all transformer to cpu. the error is from the following code:

          logits = self.transformer.forward_with_cond_scale(
              ids = sample_semantic_ids,
              text_embeds = text_embeds,
              cond_scale = cond_scale,
              **kwargs
          )

[jinyuli]

I tried again today, it's able to generate audio with CUDA now, I didn't change any thing. thanks for the great work.

[lucidrains]

that's what i like to hear! you are welcome! 😄