Module quantized_metavoice 

Quantized MetaVoice model implementation.

MetaVoice is a conditional text-to-speech model based on a transformer architecture. This implementation provides quantization for reduced memory and compute.

Key characteristics:

• Transformer-based autoregressive decoder
• Speaker conditioning
• Support for 8-bit quantization
• Key-value caching for efficient inference
• RMS normalization layers

References:

• MetaVoice Code

Re-exports

pub use crate::quantized_var_builder::VarBuilder;

Modules

transformer