SAM Audio needs about 6.1 GB of memory at Q8_0 (a 5.0 GB file) with an 8K context window, so it fits on a 8 GB GPU. This repository has 9 quants, from 2.4 GB (Q8_0) to 15.4 GB (F32).
SAM Audio – audio separation model in GGUF format
SAM Audio is a GGUF model for separating sound described by text, masked images, or time spans from an input recording. It includes SAM weights, a T5 text encoder, tokenizer, and configuration for native inference with audio.cpp.
Good at
- Audio source separation
- Text‑driven audio editing
- Masked image or video prompts
Limits
- No vision support beyond masked images/videos
- Requires CUDA/Vulkan/Metal runtime
- Large VRAM needed for higher precision
Best for
- Podcast audio cleanup
- Music remixing
- Audio editing with visual cues
Who it’s for. Developers and researchers needing a lightweight audio separation tool that runs natively on audio.cpp. Suitable for projects that require text or visual prompts to isolate audio sources.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q8_0 | 2.4 GB | 3.2 GB | Fits | Get file → | |
| BF16 | 2.8 GB | 3.7 GB | Fits | Get file → | |
| Q8_0 | 3.1 GB | 4.0 GB | Fits | Get file → | |
| BF16 | 4.1 GB | 5.1 GB | Fits | Get file → | |
| Q8_0 | 5.0 GB | 6.1 GB | Fits | Get file → | |
| F32 | 5.6 GB | 6.8 GB | Fits | Get file → | |
| BF16 | 7.7 GB | 9.1 GB | Too big | Get file → | |
| F32 | 8.3 GB | 9.8 GB | Too big | Get file → | |
| F32 | 15.4 GB | 17.7 GB | Too big | Get file → |
For 8 GB VRAM, use the Q8_0 2.38 GB file (≈3.2 GB RAM). For 12 GB VRAM, the BF16 2.82 GB file (≈3.7 GB RAM) is a good balance. If more memory is available, the F32 5.64 GB file (≈6.8 GB RAM) offers full precision.
What GPU do you need for SAM Audio?
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 Q8_0 (3.1 GB file, about 4.0 GB in use).
- 6 GB VRAM: the best quant that fits is BF16 (4.1 GB file, about 5.1 GB in use).
- 8 GB VRAM: the best quant that fits is F32 (5.6 GB file, about 6.8 GB in use).
- 12 GB VRAM: the best quant that fits is F32 (8.3 GB file, about 9.8 GB in use).
- 16 GB VRAM: the best quant that fits is F32 (8.3 GB file, about 9.8 GB in use).
- 24 GB VRAM: the best quant that fits is F32 (15.4 GB file, about 17.7 GB in use).
Which SAM Audio quant should you choose?
Q8_0 (5.0 GB) is the middle option in this repository. Close to lossless. It costs roughly twice the memory of a 4-bit file, so it suits cases where you have plenty of VRAM.
If memory is tight, the smallest file is Q8_0 at 2.4 GB, with the quality trade-offs that come with fewer bits. Close to lossless. It costs roughly twice the memory of a 4-bit file, so it suits cases where you have plenty of VRAM.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about SAM Audio
How much VRAM does SAM Audio need?
About 6.1 GB at Q8_0 with an 8K context window. That is the 5.0 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 SAM Audio on an 8 GB GPU?
Yes, with F32: the 5.6 GB file needs about 6.8 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of SAM Audio GGUF?
It depends on the quant: Q8_0 is 2.4 GB, BF16 is 2.8 GB, Q8_0 is 3.1 GB, BF16 is 4.1 GB, Q8_0 is 5.0 GB, F32 is 5.6 GB and 3 more.
Where can I download SAM Audio GGUF?
The files are on Hugging Face in the audio-cpp/SAM-Audio-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 SAM Audio 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 SAM Audio on 8 GB VRAM?
Yes, the Q8_0 2.38 GB file requires about 3.2 GB of VRAM at 8 k context, fitting within 8 GB.
Does SAM Audio support video input?
It accepts a masked video via reference_video_path, but video decoding needs FFmpeg/libav and only one visual option at a time.
Is the model licensed for commercial use?
The license is listed as "other"; consult the repository for specific terms before commercial deployment.