mlx-community/ACE-Step-1.5-turbo-bf16

ACE-Step 1.5 turbo (DiT + condition encoder + FSQ tokenizer / detokenizer) β€” MLX, bf16. Converted from ACE-Step/Ace-Step1.5 @ 19671f4 by the converter of xocialize/mlx-ace-step-swift (Tools/convert/convert.py, mlx 0.32.4), a full port of ace-step/ACE-Step-1.5 @ ca1e85f to Swift-MLX.

What is in the bundle

acestep-v15-turbo of ACE-Step 1.5 in MLX layout, bf16: the 2 B flow-matching DiT (24 layers, alternating sliding-128 / full attention, adaLN, cross-attention), the condition encoder (lyric encoder 8 layers, timbre encoder 4 layers, text projector), the FSQ audio tokenizer (attention pooler + ResidualFSQ, levels [8, 8, 8, 5, 5, 5]) and the 2-layer detokenizer β€” 677 tensors, keys as in the checkpoint except decoder.proj_in.1 / decoder.proj_out.1 β†’ decoder.proj_in / decoder.proj_out with the convolutions in MLX's [out, k, in] layout; plus silence_latent.safetensors ([15000, 64] fp32, the silence latent the pipeline pads and references with). config.json is the checkpoint's with an mlx_ace_step provenance block.

Licence

MIT. The Hugging Face repository ACE-Step/Ace-Step1.5 ships no LICENSE file; its model-card tag and README state MIT and the GitHub LICENSE (Copyright (c) 2026 ACEStep) is the controlling text.

The full notice travels with the bundle as NOTICE; LICENSE is the MIT text. Nothing in the weights is modified beyond the layout conversion (key prefix / conv layout / weight-norm folding) β€” the converter Tools/convert/convert.py in xocialize/mlx-ace-step-swift is gated bit-equal against upstream's own dit_convert.py / vae_convert.py (its MEASUREMENTS.md Β§V1).

Use

Every tier of the Swift package loads four bundles β€” the turbo DiT, one planner, the text encoder and the VAE β€” and this bundle is one of them. The planner choices: ACE-Step-1.5-planner-1.7B-8bit for .int8 (the default), -4bit for .int4, -bf16 for .bf16 / .fp32, and ACE-Step-1.5-planner-0.6B-bf16 for the lite planner on any tier. The package, xocialize/mlx-ace-step-swift, runs ACE-Step 1.5 on Apple Silicon through MLX Swift and plugs into MLXEngine on the music capability (caption + lyrics β†’ a 48 kHz stereo song; reference-audio timbre; cover / repaint / extend of a source clip; LRC timestamps). The package's ACEStepConfiguration names these repos and the engine materialises them; the parity gates against the upstream oracle (fp32, TF32 off) are recorded in the package's MEASUREMENTS.md.

let engine = MLXServeEngine()
let id = try await engine.register(ACEStepPackage.registration, configuration: ACEStepConfiguration())   // the .int8 default
try await engine.prepare(.music, package: id)
let song = try await engine.run(MusicRequest(caption: "…", lyrics: "[Verse]\n…", durationSeconds: 120, seed: 7), package: id)

Upstream: ACE-Step/Ace-Step1.5 @ 19671f4 (weights), ace-step/ACE-Step-1.5 @ ca1e85f (code) β€” ACE Studio + StepFun, tech report.

Downloads last month
102
Safetensors
Model size
2B params
Tensor type
BF16
Β·
MLX
Hardware compatibility
Log In to add your hardware

Quantized

Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Model tree for mlx-community/ACE-Step-1.5-turbo-bf16

Finetuned
(27)
this model

Collection including mlx-community/ACE-Step-1.5-turbo-bf16

Paper for mlx-community/ACE-Step-1.5-turbo-bf16