longcat-audiodit-3.5b-8bit-mlx

LongCat AudioDiT 3.5B — MLX 8-bit

This repository contains a self-contained MLX-native int8 conversion of LongCat AudioDiT 3.5B for local text-to-speech on Apple Silicon.

It is intended for local speech generation with mlx-speech, without a PyTorch runtime at inference time.

Model Details

• Developed by: AppAutomaton
• Shared by: AppAutomaton on Hugging Face
• Upstream model: meituan-longcat/LongCat-AudioDiT-3.5B
• Task: text-to-speech and voice cloning
• Runtime: MLX on Apple Silicon
• Precision: int8 quantized weights with bundled tokenizer

Bundle Contents

This bundle is self-contained and includes:

• config.json
• model.safetensors
• tokenizer files (tokenizer.json, tokenizer_config.json, special_tokens_map.json)

How to Get Started

Basic generation:

python scripts/generate/longcat_audiodit.py \
  --text "Hello from LongCat AudioDiT." \
  --output-audio outputs/longcat.wav

Voice cloning:

python scripts/generate/longcat_audiodit.py \
  --text "Hello from LongCat AudioDiT." \
  --prompt-text "Original speaker text." \
  --prompt-audio /path/to/prompt.wav \
  --output-audio outputs/longcat_clone.wav \
  --guidance-method apg

Minimal Python usage:

from pathlib import Path

from mlx_speech.generation.longcat_audiodit import generate_longcat_audiodit

result = generate_longcat_audiodit(
    text="Hello from LongCat AudioDiT.",
    output_audio="outputs/longcat.wav",
)

Notes

• This repo contains the quantized MLX runtime artifact only.
• The conversion preserves the LongCat AudioDiT diffusion transformer and bundled VAE for waveform decode.
• Voice cloning uses --guidance-method apg (Adaptive Projected Guidance) or cfg (Classifier-Free Guidance, default). --guidance-strength controls speaker adherence (default: 4.0).
• The current bundle is intended for local MLX runtime use and parity validation.

Links

• Source code: mlx-speech
• Project page: appautomaton.com/mlx-speech
• More examples: AppAutomaton

License

MIT License — following the upstream license published with meituan-longcat/LongCat-AudioDiT-3.5B.