how to run this code?
i am getting errors only. when i am using safetensor model file i am getting hyper parameter error, if i am running your code without doing any changes getting
NotImplementedError: Cannot copy out of meta tensor; no data! Please use torch.nn.Module.to_empty() instead of torch.nn.Module.to() when moving module from meta to a different device.
your sample code not running as it is, any changes needs to do ?
pip install transformers==4.49.0 pydub soundfile safetensors huggingface_hub , this will fix you issue
.cache\huggingface\modules\transformers_modules\ai4bharat\IndicF5\b82d286220e3070e171f4ef4b4bd047b9a447c9a
ERROR: Failed to load model. Please check internet connection, Hugging Face login, and model access. Error: load_model() missing 1 required positional argument: 'ckpt_path'
after installed the above packages, i am getting the above error in the model.py file
now i can able to run and get the output but the audio is some noise only.
hey
@ERDllm
, I give you a Colab file like https://colab.research.google.com/drive/1OXG2R3-AFGTcST5nucTWoUq-ZQvtel0a?usp=sharing. In the file you have easily run the model. you must try once
hey @ERDllm , I give you a Colab file like
https://colab.research.google.com/drive/1OXG2R3-AFGTcST5nucTWoUq-ZQvtel0a?usp=sharing. In the file you have easily run the model. you must try once
Hi RedRepter,
While I got build up without errors on windows with i9+Rtx3080 Gpu,
Only random garbled fragments are audible and with each run for same input text, some garbled short audio is generated.
Tried with exact reference audio from the Indicf5 prompts and the exact reference text.
Also tried 2 different language input text s but the garbled short audio generation is NOT fixed.
Could you please let us know in case you used this Model for indian language voice generation, did it work?
If this model not maintained now?
hey @ERDllm , I give you a Colab file like
https://colab.research.google.com/drive/1OXG2R3-AFGTcST5nucTWoUq-ZQvtel0a?usp=sharing. In the file you have easily run the model. you must try onceHi RedRepter,
While I got build up without errors on windows with i9+Rtx3080 Gpu,
Only random garbled fragments are audible and with each run for same input text, some garbled short audio is generated.
Tried with exact reference audio from the Indicf5 prompts and the exact reference text.Also tried 2 different language input text s but the garbled short audio generation is NOT fixed.
Could you please let us know in case you used this Model for indian language voice generation, did it work?
If this model not maintained now?
Hey @k1-m ,
For your inquiry. I encountered a similar issue when attempting to load directly from the transform. My solution involved creating a FastAPI application with minor modifications, which resolved the problem for me.
Regarding device limitations, I was unable to load this model on my local machine due to its specifications (i5 9th gen, RTX 1650ti). Instead, I successfully deployed the model on the Modal platform. I am currently unfamiliar with the maintenance aspects of this setup.
Thanks for the reply,
Coming to your note "My solution involved creating a FastAPI application with minor modifications", does it need subscription to run this modal from a online server and acces with FastAPI?
It would be very helpful if I can generate voice on windows setup (PC met the CPU+GPU requirement)
(Currently it emits halucinated 2 seconds audio unfortunately, and the generated sounds keep changing with each run even with same unchanged fixed input text)
@k1-m
No, in Modal ( https://modal.com )you only need an account for sign-up, and then for the first month it gives you 5 dollars. With this you can deploy your ML project with the FastAPI GPU, and after the first month it gives you, I think, 30 dollars, so that is enough for this project. And also, I already made this model API, so if you want, DM me in Discord ( devilxgamer221 ) by the way its free , and I will send you the GitHub link.