Instructions to use nvidia/personaplex-7b-v1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Moshi
How to use nvidia/personaplex-7b-v1 with Moshi:
# pip install moshi # Run the interactive web server python -m moshi.server --hf-repo "nvidia/personaplex-7b-v1" # Then open https://localhost:8998 in your browser
# pip install moshi import torch from moshi.models import loaders # Load checkpoint info from HuggingFace checkpoint = loaders.CheckpointInfo.from_hf_repo("nvidia/personaplex-7b-v1") # Load the Mimi audio codec mimi = checkpoint.get_mimi(device="cuda") mimi.set_num_codebooks(8) # Encode audio (24kHz, mono) wav = torch.randn(1, 1, 24000 * 10) # [batch, channels, samples] with torch.no_grad(): codes = mimi.encode(wav.cuda()) decoded = mimi.decode(codes) - Inference
- Notebooks
- Google Colab
- Kaggle
Modal example (serverless GPU)
#62
by moeadham - opened
Hey, if anyone is looking to try this out themselves, feel free to try:
https://github.com/moeadham/personaplex-modal
It deploys the model and demo app to modal, uses memory snapshotting and gets you a functional demo quickly and with low cost.
Overall, I found that the quality of the model in reality was quite different from the demos selected. While it is okay, I think it needs something like KAME or another finetune to keep it rational.
Also the voice quality is quite disappointing if you are used to current SOTA TTS models.
Cheers