RuntimeError in loading state_dict when calling CLVPMetric(device='cuda')

[a1rishav]

Error details:

RuntimeError: Error(s) in loading state_dict for CLVP:
   Missing key(s) in state_dict: "text_pos_emb.weight", "text_transformer.layers.layers.0.0.scale" ....
   .....  mismatch for to_speech_latent.weight: copying a param with shape torch.Size([1024, 1024]) from checkpoint, the shape in current model is torch.Size([512, 512]).

[neonbjb]

Ah apologies - this repo uses the original CLVP: https://huggingface.co/jbetker/tortoise-tts-v2/blob/main/.models/clvp.pth

It would be better if it were updated to use the bigger one.

[a1rishav]

Thanks for getting back, I am still getting the same error with the older clvp.

[neonbjb]

That doesn't seem right; the error should at least be different (CLVP1 has d_model=512, CLVP2 has d_model=1024)

…

On Mon, Aug 28, 2023 at 10:47 PM Rishav Kumar ***@***.***> wrote: Thanks for getting back, I am still getting the same error with the older clvp. — Reply to this email directly, view it on GitHub <#6 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAGLMOSFX3PZH6WLJRDXTKDXXVX6RANCNFSM6AAAAAA4BJRRAI> . You are receiving this because you commented.Message ID: ***@***.***>

-- - James Betker

[a1rishav]

Yeah, you are right, the current error isn't exactly same as previous, I meant that it is still about the size mismatch
RuntimeError: Error(s) in loading state_dict for CLVP:

Missing key(s) in state_dict: "text_pos_emb.weight", "text_transformer.layers.layers.0.0.scale", "text_transformer.layers.layers.0.0.fn.norm.weight",  .......
......size mismatch for text_emb.weight: copying a param with shape torch.Size([256, 512]) from checkpoint, the shape in current model is torch.Size([148, 512]).

[WhiteTeaDragon]

Has this issue been solved? I get the same mistake as the previous commenter