Stable Audio 3 Small MLX
Powered by Stability AI.
Unmodified official optimized MLX tensors for Music and SFX. Source: https://huggingface.co/stabilityai/stable-audio-3-optimized/tree/da6edc54ddba10bfd79a077102ded687f80e882b . Upstream inference code: https://github.com/Stability-AI/stable-audio-3/tree/3a82c807b69cf4b7c5c05270011a5d5e47abac18 .
AI2Apps distributes this snapshot for non-commercial installation, evaluation and testing. Users must accept the Stability AI Community License and Gemma Terms before AI2Apps downloads these weights. Commercial use is governed by the original terms, including applicable registration and separate-license requirements. See LICENSE.md, LICENSE_GEMMA.md and Notice. The original terms and incorporated use restrictions apply to every user.
Music uses MLX/dit_sm-music_f16.npz; SFX uses MLX/dit_sm-sfx_f16.npz. Both require MLX/t5gemma_f16.npz and MLX/same_s_decoder_f32.npz. Text-only generation does not require the audio encoder. No tensor conversion or modification was performed.