Crash when launching new model

[ApatheticWrath]

Self Checks

• This is only for bug report, if you would like to ask a question, please head to Discussions.
• I have searched for existing issues search for existing issues, including closed ones.
• I confirm that I am using English to submit this report (我已阅读并同意 Language Policy).
• [FOR CHINESE USERS] 请务必使用英文提交 Issue，否则会被关闭。谢谢！:）
• Please do not modify this template :) and fill in all the required fields.

Cloud or Self Hosted

Self Hosted (Source)

Steps to reproduce

I click start.bat and go to inference configuration.
for Decoder Model Path path i select:
checkpoints\fish-speech-1.4\firefly-gan-vq-fsq-8x1024-21hz-generator.pth
Decoder Model Config:
firefly_gan_vq
LLAMA Model Path:
checkpoints\fish-speech-1.4
Compile model:
No

click Open Inference Server

I get this

"HF_ENDPOINT: https://huggingface.co"
"NO_PROXY: "
model.pth 已存在，跳过下载。
README.md 已存在，跳过下载。
special_tokens_map.json 已存在，跳过下载。
tokenizer_config.json 已存在，跳过下载。
tokenizer.json 已存在，跳过下载。
config.json 已存在，跳过下载。
firefly-gan-vq-fsq-4x1024-42hz-generator.pth 已存在，跳过下载。
ffmpeg.exe 已存在，跳过下载。
ffprobe.exe 已存在，跳过下载。
asr-label-win-x64.exe 已存在，跳过下载。
Debug: flags =

Next launch the page...
['', 'C:\\Users\\J__mo\\Downloads\\fish-speech-1.2.1\\fish-speech-1.2.1\\fish_speech\\webui', 'C:\\Users\\J__mo\\Downloads\\fish-speech-1.2.1\\fish-speech-1.2.1', 'C:\\Users\\J__mo\\Downloads\\fish-speech-1.2.1\\fish-speech-1.2.1\\fishenv\\env\\python310.zip', 'C:\\Users\\J__mo\\Downloads\\fish-speech-1.2.1\\fish-speech-1.2.1\\fishenv\\env\\DLLs', 'C:\\Users\\J__mo\\Downloads\\fish-speech-1.2.1\\fish-speech-1.2.1\\fishenv\\env\\lib', 'C:\\Users\\J__mo\\Downloads\\fish-speech-1.2.1\\fish-speech-1.2.1\\fishenv\\env', 'C:\\Users\\J__mo\\Downloads\\fish-speech-1.2.1\\fish-speech-1.2.1\\fishenv\\env\\lib\\site-packages', '__editable__.fish_speech-0.1.0.finder.__path_hook__']
You are in  C:\Users\J__mo\Downloads\fish-speech-1.2.1\fish-speech-1.2.1
Running on local URL:  http://127.0.0.1:7860

To create a public link, set `share=True` in `launch()`.
2024-09-11 20:50:33.852 | INFO     | __main__:clean_infer_cache:146 - C:\Users\J__mo\AppData\Local\Temp\gradio was not found
2024-09-11 20:50:49.634 | INFO     | __main__:<module>:584 - Loading Llama model...
2024-09-11 20:50:55.998 | INFO     | tools.llama.generate:load_model:352 - Restored model from checkpoint
2024-09-11 20:50:55.999 | INFO     | tools.llama.generate:load_model:356 - Using DualARTransformer
2024-09-11 20:50:56.001 | INFO     | __main__:<module>:591 - Llama model loaded, loading VQ-GAN model...
Traceback (most recent call last):
  File "C:\Users\J__mo\Downloads\fish-speech-1.2.1\fish-speech-1.2.1\tools\webui.py", line 593, in <module>
    decoder_model = load_decoder_model(
  File "C:\Users\J__mo\Downloads\fish-speech-1.2.1\fish-speech-1.2.1\tools\vqgan\inference.py", line 41, in load_model
    result = model.load_state_dict(state_dict, strict=False)
  File "C:\Users\J__mo\Downloads\fish-speech-1.2.1\fish-speech-1.2.1\fishenv\env\lib\site-packages\torch\nn\modules\module.py", line 2191, in load_state_dict
    raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for FireflyArchitecture:
        size mismatch for quantizer.residual_fsq.rvqs.0.project_in.weight: copying a param with shape torch.Size([4, 64]) from checkpoint, the shape in current model is torch.Size([4, 128]).
        size mismatch for quantizer.residual_fsq.rvqs.0.project_out.weight: copying a param with shape torch.Size([64, 4]) from checkpoint, the shape in current model is torch.Size([128, 4]).
        size mismatch for quantizer.residual_fsq.rvqs.0.project_out.bias: copying a param with shape torch.Size([64]) from checkpoint, the shape in current model is torch.Size([128]).
        size mismatch for quantizer.residual_fsq.rvqs.1.project_in.weight: copying a param with shape torch.Size([4, 64]) from checkpoint, the shape in current model is torch.Size([4, 128]).
        size mismatch for quantizer.residual_fsq.rvqs.1.project_out.weight: copying a param with shape torch.Size([64, 4]) from checkpoint, the shape in current model is torch.Size([128, 4]).
        size mismatch for quantizer.residual_fsq.rvqs.1.project_out.bias: copying a param with shape torch.Size([64]) from checkpoint, the shape in current model is torch.Size([128]).
        size mismatch for quantizer.residual_fsq.rvqs.2.project_in.weight: copying a param with shape torch.Size([4, 64]) from checkpoint, the shape in current model is torch.Size([4, 128]).
        size mismatch for quantizer.residual_fsq.rvqs.2.project_out.weight: copying a param with shape torch.Size([64, 4]) from checkpoint, the shape in current model is torch.Size([128, 4]).
        size mismatch for quantizer.residual_fsq.rvqs.2.project_out.bias: copying a param with shape torch.Size([64]) from checkpoint, the shape in current model is torch.Size([128]).
        size mismatch for quantizer.residual_fsq.rvqs.3.project_in.weight: copying a param with shape torch.Size([4, 64]) from checkpoint, the shape in current model is torch.Size([4, 128]).
        size mismatch for quantizer.residual_fsq.rvqs.3.project_out.weight: copying a param with shape torch.Size([64, 4]) from checkpoint, the shape in current model is torch.Size([128, 4]).
        size mismatch for quantizer.residual_fsq.rvqs.3.project_out.bias: copying a param with shape torch.Size([64]) from checkpoint, the shape in current model is torch.Size([128]).

The 1.2 model loads and works fine just not 1.4

✔️ Expected Behavior

For the inference server to launch with the new model.

❌ Actual Behavior

It crashed.

[PoTaTo-Mika]

Try the command with your virtual environment activated.
python tools/webui.py
please run the command under the fish-speech folder.
Also, please ensure your code version, you can't run the V1.4 model with V1.2 source code.

[ApatheticWrath]

I just grabbed latest release without noticing the date. I wasn't using latest repo code, it works now. Thank you.