Module mimi 

Mimi neural audio tokenizer support. Mimi neural audio tokenizer support.

Mimi is the neural audio codec used by Moshi-family realtime models. This module implements backend-neutral checkpoint parameters, the split residual vector quantizer, and the non-streaming SEANet/transformer encoder and decoder used to map between PCM and Mimi codebook tokens.

Structs

Config

Mimi codec configuration.

Conv1x1NoBias

Bias-free 1x1 convolution over [batch, channels, frames] tensors.

EuclideanCodebook

Euclidean codebook backed by EMA cluster statistics.

Mimi

Mimi audio tokenizer.

MimiParameterRequirement

Exact released-checkpoint requirement for one Mimi parameter.

PreparedMimiArtifact

Header-admitted Mimi artifact with an immutable exact parameter plan.

ResidualVectorQuantization

Residual vector quantization layers.

ResidualVectorQuantizer

Residual vector quantizer branch.

SelectedMimiArtifact

Backend-admitted Mimi plan ready for native materialization.

SplitResidualVectorQuantizer

Split residual vector quantizer used by Mimi.

VectorQuantization

Single vector-quantization layer.

Enums

MimiArtifactError

Failure while preparing a released Mimi artifact from neutral metadata.

MimiConstructionError

Failure while selecting, materializing, or constructing Mimi on a backend.

ResampleMethod

Mimi resampling strategy.

Functions

construct

Selects generic backend mechanisms, then constructs Mimi through the exact neutral artifact plan.

prepare_checkpoint

Inspects and admits an exact released Mimi SafeTensors artifact without reading tensor payloads.

prepare_source

Admits an already opened backend-neutral SafeTensors-compatible source. No tensor payload is acquired during preparation.

released_checkpoint_requirements

Returns the complete released-checkpoint schema derived from Mimi’s authoritative parameter topology without inspecting an artifact.