thepatch/stable-audio-open-1.0-GGUF

Stable Audio Open 1.0 GGUF

GGUF conversions of stabilityai/stable-audio-open-1.0 for sa3.cpp's optional native stable-audio-tools runtime.

Each encoding is a self-contained DiT, 128-token T5-base encoder, and Oobleck decoder bundle. The T5 and Oobleck artifacts are byte-identical to the files in the Foundation-1 GGUF repository.

Tier	Complete bundle	Guidance
F16	2,377 MiB	reference
Q8_0	1,296 MiB	conservative high fidelity
Q5_K_M	922 MiB	compact
Q4_K_M	832 MiB	smallest footprint

All quantized tensors passed the per-tensor 0.990 cosine gate. Matched CUDA renders measured envelope/log-magnitude cosine of 0.9959/0.9784 at Q8, 0.9021/0.8742 at Q5, and 0.9071/0.8789 at Q4. All four tiers passed listening tests; Q8 remains the conservative recommendation for this model.

The official inference profile is DPM++ 3M SDE, sigma 0.3–500, 100 steps, and CFG 7. Stable Audio Open 1.0 supports up to approximately 47 seconds at 44.1 kHz stereo.

cmake -S . -B build-sat -DSA3_BUILD_SAT=ON -DSA3_CUDA=ON
cmake --build build-sat --config Release --target sat-generate
python tools/download_models.py --sat --sat-model stable-audio-open-1.0
$env:SA3_DEVICE="cuda"
sat-generate --model stable-audio-open-1.0 --prompt "cinematic ambient soundscape" `
  --seconds 47 --seed 42 --out sao1.wav

Sources and licensing

• Stable Audio Open 1.0 source revision: f21265c1e2710b3bd2386596943f0007f55f802e.
• T5-base revision: a9723ea7f1b39c1eae772870f3b547bf6ef7e6c1.

The Stable Audio weights remain under the Stability AI Community License included as LICENSE.md. T5-base is covered by LICENSE_T5.md. See NOTICE for conversion details. These artifacts rename tensors, serialize them as GGUF, and quantize selected tensors; they do not retrain the model.

Powered by Stability AI.