The source float32 tensors were resharded from the original 7 near-5GB shards into 14 smaller shards for more reliable Hugging Face upload. Tensor dtypes and values are unchanged.

Load

python

from mimo_mlx.asr import MiMoASR

asr = MiMoASR(
    asr_model_dir="models/MiMo-V2.5-ASR",
    audio_tokenizer_dir="models/MiMo-Audio-Tokenizer",
)
text = asr.transcribe("audio.wav", language="zh")

The audio tokenizer is required separately. Use the official XiaomiMiMo/MiMo-Audio-Tokenizer files or the local directory expected by the project.

Source

Base model: https://huggingface.co/XiaomiMiMo/MiMo-V2.5-ASR

carloshuang1224/MiMo-V2.5-ASR-MLX-F32

MiMo-V2.5-ASR MLX F32

Files

Load

Source

carloshuang1224/
MiMo-V2.5-ASR-MLX-F32