GGUF conversions of RoyalCities/Foundation-1 for sa3.cpp's optional native stable-audio-tools runtime. Foundation-1 is a full finetune of Stable Audio Open 1.0 and uses the same T5-base and Oobleck architecture. Each encoding is a complete three-file bundle: All quantized tensors passed the per-tensor 0.990 cosine gate. Matched CUDA renders measured envelope/log-magnitude cosine of 0.9967/0.9901 at Q8, 0.9904/0.9550 at Q5, and 0.9643/0.9284 at Q4. All four tiers passed listening tests. Foundation-1 was trained for 4- or 8-bar clips at 100, 110, 120, 128, 130, 140, or 150 BPM. The prompt should include the selected bar count and BPM. sa3.cpp's Foundation profile calculates the exact sample…
Organization
The Collabage Patch
thepatch
currently focused on musicgen until stable audio weights emerge.
Models in Library1
Datasets in Library0
Models on Hugging Face62
Followers18
Compare the published evidence
This profile connects the 1 model and dataset record shown below to their upstream cards and sources. The library counts describe SAVRN's coverage, rather than this publisher's entire output.
Reported licenses: other. Open a record to review its specific license and conditions.
Compare model specifications, source dates, and hardware estimates on the individual records. Publisher identity verification is separate from a model's quality, deployment readiness, or suitability for your work.