stable-audio-3-medium SAME-L needs about 1.2 GB of memory at Q4_K_M (a 0.6 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 11 quants, from 0.0 GB (F32) to 5.8 GB (F32).
Stable Audio 3 Medium in GGUF format for sa3.cpp
Stable Audio 3 Medium is converted to GGUF for use with sa3.cpp, a portable C++/GGML port. The model runs on CPU, CUDA, Vulkan, or Metal and is validated against the PyTorch reference. It includes DiT, SAME, conditioner, encoder, and tokenizer components.
Good at
- Audio generation
- Fast inference on GPU
- Cross‑platform compatibility
Limits
- No vision support
- Requires separate encoder/tokenizer
- Large memory for high precision
Best for
- Short audio clips generation
- Batch audio synthesis
- Low‑latency audio inference
Who it’s for. Developers and researchers needing a lightweight, portable audio generation model that can run on diverse hardware. Suitable for integration into custom audio pipelines or embedded systems.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| F32 | 0.0 GB | 0.5 GB | Fits | Get file → | |
| Q4_K_M | 0.6 GB | 1.2 GB | Fits | Get file → | |
| Q5_K_M | 0.6 GB | 1.2 GB | Fits | Get file → | |
| Q8_0 | 0.9 GB | 1.5 GB | Fits | Get file → | |
| Q4_K_M | 1.0 GB | 1.6 GB | Fits | Get file → | |
| Q5_K_M | 1.1 GB | 1.7 GB | Fits | Get file → | |
| Q8_0 | 1.6 GB | 2.2 GB | Fits | Get file → | |
| F16 | 1.7 GB | 2.4 GB | Fits | Get file → | |
| F16 | 2.9 GB | 3.8 GB | Fits | Get file → | |
| F32 | 3.4 GB | 4.3 GB | Fits | Get file → | |
| F32 | 5.8 GB | 7.0 GB | Fits | Get file → |
For 8 GB VRAM, use Q4_K_M (0.6 GB, ~1.2 GB RAM). For 4 GB VRAM, use Q4_K_M (1.01 GB, ~1.6 GB RAM). Higher precision (F16, F32) requires more memory and is best on GPUs with 12 GB+ VRAM.
What GPU do you need for stable-audio-3-medium SAME-L?
These figures add the file size, an fp16 KV cache at an 8K context window and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: the best quant that fits is F16 (2.9 GB file, about 3.8 GB in use).
- 6 GB VRAM: the best quant that fits is F32 (3.4 GB file, about 4.3 GB in use).
- 8 GB VRAM: the best quant that fits is F32 (5.8 GB file, about 7.0 GB in use).
Which stable-audio-3-medium SAME-L quant should you choose?
Q4_K_M is the usual starting point for stable-audio-3-medium SAME-L: it balances file size and quality well. Here it is a 0.6 GB file.
The Q8_0 file is 0.9 GB, about 1.5 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.
If memory is tight, the smallest file is F32 at 0.0 GB, with the quality trade-offs that come with fewer bits. The unquantized model. It gives original quality and needs the most memory, so it is mainly for conversion or reference.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about stable-audio-3-medium SAME-L
How much VRAM does stable-audio-3-medium SAME-L need?
About 1.2 GB at Q4_K_M with an 8K context window. That is the 0.6 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run stable-audio-3-medium SAME-L on an 8 GB GPU?
Yes, with F32: the 5.8 GB file needs about 7.0 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of stable-audio-3-medium SAME-L GGUF?
It depends on the quant: F32 is 0.0 GB, Q4_K_M is 0.6 GB, Q5_K_M is 0.6 GB, Q8_0 is 0.9 GB, Q4_K_M is 1.0 GB, Q5_K_M is 1.1 GB and 5 more.
Where can I download stable-audio-3-medium SAME-L GGUF?
The files are on Hugging Face in the thepatch/stable-audio-3-medium-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does stable-audio-3-medium SAME-L use?
Hugging Face lists the license as other. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run Stable Audio 3 Medium on 8 GB VRAM?
Yes, using Q4_K_M or Q5_K_M quantizations keeps VRAM usage below 8 GB and runs efficiently on most GPUs.
Does this model support vision inputs?
No, the model has no vision capabilities and only processes audio data.
What is the fastest quantization for CUDA?
CUDA gains roughly 33% speed with Q4_K_M, Q5_K_M, or Q8_0 compared to F16, but the exact improvement depends on the GPU.