MODEL UPLOADING ERROR.

[sudden bone]

HEllo , I've been trying to upload my rvc model and index file in w-okada but it doesn't appear And the Other model files are appearing.

[wide comet]

show a screenshot of the error, of your wokada, and also the model downlload link

[sudden bone]

Now it appears but when i start the voice changer, in my terminal it gives error like The following operation failed in the TorchScript interpreter.
Traceback of TorchScript (most recent call last):
RuntimeError: Expected size for first two dimensions of batch2 tensor to be: [1, 256] but got: [1, 768].
Traceback (most recent call last):
File "voice_changer\VoiceChangerManager.py", line 212, in change_voice
audio, vol, perf = self.vc.on_request(receivedData)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "torch\utils_contextlib.py", line 116, in decorate_context
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "voice_changer\VoiceChangerV2.py", line 162, in on_request
result, vol = self.process_audio(audio_in)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "voice_changer\VoiceChangerV2.py", line 126, in process_audio
audio, vol = self.vcmodel.inference(audio_in)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "voice_changer\RVC\RVCr2.py", line 239, in inference
self.pipeline.exec(
File "voice_changer\RVC\pipeline\Pipeline.py", line 235, in exec
out_audio = self.inferencer.infer(feats, p_len, pitch, pitchf, sid, skip_head, return_length, formant_length).float()
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "voice_changer\RVC\inferencer\RVCInferencerv2.py", line 60, in infer
res = self.model.infer(
^^^^^^^^^^^^^^^^^
RuntimeError: The following operation failed in the TorchScript interpreter.
Traceback of TorchScript (most recent call last):

[sudden bone]

Here is my model link https://drive.google.com/file/d/1JocWG2A6KVUz6O9GgqXi_9o2klT4GJGl/view?usp=drive_link

[wide comet]

!give-media-perms 1h @sudden bone

show also a screenshot of what voice changer you're using

you could be using a 1 yeard old version got off youtube

[wide comet]

the files seem weird, it contains no .pth and has other useless files

you deffo are using something which is years old off youtube

[sudden bone]

I have used https://huggingface.co/Shadicti/deiteris-Fork/blob/main/voice-changer-windows-nvidia-b2332.zip this version

[wide comet]

yup that's right then, i guess you read https://rentry.co/forkvoicechangerguide

did you make that model? or did u find it randomly?

if you made it, what program did u use?

[sudden bone]

I'm using safetensors file instead pth. In statting i run pth but the model does not appear in w-okada

[sudden bone]

I made

[sudden bone]

Rvc webui official local host

[wide comet]

so, you mean rvc mainline?

[sudden bone]

https://huggingface.co/lj1995/VoiceConversionWebUI/resolve/main/RVC1006Nvidia.7z This is the project i ran it

[wide comet]

yup we commonly call that mainline since there's also other rvc forks

[sudden bone]

Oh

[wide comet]

I see, wokada deiteris fork supports safetensors too iirc

did you upload only the safetensors and added index in wokada?

you don't need the other files for inference

[sudden bone]

Yeah.i uploaded safetensor and index. But actually in my training folder there's pth file too but i didn't use it

[wide comet]

did u use tensorboard to train the model?

i'm not aware of a safetensors option for rvc mainline training tbh, it sounds weird

[sudden bone]

No what's that

[wide comet]

https://docs.aihub.gg/rvc/resources/training/#tensorboard

[sudden bone]

The main reason to run safetensors is, i get this error while running pth file, Traceback (most recent call last):
File "restapi\MMVC_Rest_Fileuploader.py", line 67, in post_load_model
File "voice_changer\VoiceChangerManager.py", line 118, in load_model
slotInfo = RVCModelSlotGenerator.load_model(params)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "voice_changer\RVC\RVCModelSlotGenerator.py", line 42, in load_model
slotInfo = cls._setInfoByPytorch(modelPath, slotInfo)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "voice_changer\RVC\RVCModelSlotGenerator.py", line 58, in _setInfoByPytorch
config_len = len(cpt["config"])
~~~^^^^^^^^^^
KeyError: 'config'

[wide comet]

which pth did u use?

[sudden bone]

I didn't understand. 🤔

Oh

The file starts with G

[chilly barn]

KeyError: 'config' = you're using not a voice model

so-vits likely

[sudden bone]

🤔

[chilly barn]

G_xxxx.pth is a generator weight in RVC

[sudden bone]

So i have to use D file?

[chilly barn]

D_xxxx.pth is a discriminator weight

[sudden bone]

Same happens

[chilly barn]

neither is a voice model

[sudden bone]

Lol

[chilly barn]

this is an index (usually generated at the end of the training) and a bunch of voice models saved every 20 epochs during training

[sudden bone]

In my logs, i can see only multiple G,D files that are saved in every 40 epoch

[chilly barn]

that's okay

I guess you did not check the box to save the model together with D/G saves

[sudden bone]

I see smiley_e1000 in weight folder

[chilly barn]

okay, so that's the final 1000 epoch model which is likely trash

[sudden bone]

I checked to save every epoch at wright

[sudden bone]

Hain 🙄

[chilly barn]

what are you using?

applio or mainline rvc or something else?

[sudden bone]

Rvc Web UI mainline

[chilly barn]

saves only one copy of D/G weights

saves the voice model at every save epoch together with D/G

[sudden bone]

Wait

Working

[chilly barn]

so check assets/weights

[sudden bone]

I see smiley.pth in my weight folder

And it's working

[chilly barn]

if you did not check tensorboard for the best epoch and there's only one copy, it is most likely overtrained and not the best

[sudden bone]

I checked it

Found it. Thx