hibiki-1b-mlx-q8 / README.md
anquachdev's picture
Add Hugging Face model metadata
417bed8 verified
|
Raw
History Blame Contribute Delete
433 Bytes
---
license: cc-by-4.0
language:
- fr
- en
base_model: kyutai/hibiki-1b-mlx-bf16
base_model_relation: quantized
tags:
- hibiki
- mlx
- speech-to-speech
- quantized
---
# Hibiki 1B MLX Q8
Weight-only quantization of the language model from the artifact revision `unknown`.
- Quantization: Q8, group size 64
- Quantized modules: compatible MLX Linear layers only
- Unchanged: Mimi codec, embeddings, normalisation layers, tokenizer