Instructions to use tinytrashlabs/dia2-2b-mlx-8bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use tinytrashlabs/dia2-2b-mlx-8bit with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] huggingface-cli download --local-dir dia2-2b-mlx-8bit tinytrashlabs/dia2-2b-mlx-8bit
- Dia2
How to use tinytrashlabs/dia2-2b-mlx-8bit with Dia2:
from dia2 import Dia2, GenerationConfig, SamplingConfig dia = Dia2.from_repo("tinytrashlabs/dia2-2b-mlx-8bit", device="cuda", dtype="bfloat16") config = GenerationConfig( cfg_scale=2.0, audio=SamplingConfig(temperature=0.8, top_k=50), use_cuda_graph=True, ) result = dia.generate("[S1] Hello Dia2!", config=config, output_wav="hello.wav", verbose=True) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
dia2-2b-mlx-8bit
MLX conversion of nari-labs/Dia2-2B (8bit).
Dia2 is a streaming two-speaker dialogue TTS model by Nari Labs. English only, two speakers, up to two minutes per generation. Audio is decoded through the Kyutai Mimi codec at 24 kHz.
Converted with Tools/convert-dia2.py from
TinyTrashLabs/mlx-audio-swift
and consumed by Gloam Voice Studio.
Licensed Apache 2.0, as is the source model. Third-party assets (the Mimi codec) retain their own licenses.
- Downloads last month
- -
Model size
2B params
Tensor type
U32
·
BF16 ·
Hardware compatibility
Log In to add your hardware
Quantized
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for tinytrashlabs/dia2-2b-mlx-8bit
Base model
nari-labs/Dia2-2B