Moonshine Streaming Small needs about 0.8 GB of memory at F16 (a 0.3 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 3 quants, from 0.2 GB (Q8_0) to 0.6 GB (F32).
Lightweight 0.14B ASR model for low‑latency English speech transcription
Moonshine Streaming Small is a sequence‑to‑sequence ASR model with a sliding‑window Transformer encoder and an autoregressive decoder. It is designed for on‑device, low‑latency English speech transcription on memory‑constrained hardware. The model uses a 50 Hz audio frontend and an ergodic encoder with bounded local attention.
Good at
- Low‑latency streaming ASR
- On‑device inference
- English speech transcription
Limits
- No vision or translation support
- Not evaluated for non‑English languages
- Requires 16 kHz mono WAV input
Best for
- Live captioning on edge devices
- Voice command recognition
- Real‑time transcription apps
Who it’s for. Developers building low‑resource speech‑to‑text solutions on edge hardware. The model is suitable for mobile or embedded systems that need quick, accurate English transcription without large memory footprints.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q8_0 | 0.2 GB | 0.7 GB | Fits | Get file → | |
| F16 | 0.3 GB | 0.8 GB | Fits | Get file → | |
| F32 | 0.6 GB | 1.1 GB | Fits | Get file → |
For 8 GB VRAM, the Q8_0 quant (0.2 GB) uses about 0.7 GB at 8 k context. The F16 quant (0.28 GB) uses about 0.8 GB. For 4 GB VRAM, only Q8_0 is feasible. Larger models like F32 (0.56 GB) require more than 1 GB memory and are not recommended for tight budgets.
What GPU do you need for Moonshine Streaming Small?
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 F32 (0.6 GB file, about 1.1 GB in use).
Which Moonshine Streaming Small quant should you choose?
F16 (0.3 GB) is the middle option in this repository. The unquantized model. It gives original quality and needs the most memory, so it is mainly for conversion or reference.
If memory is tight, the smallest file is Q8_0 at 0.2 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 Moonshine Streaming Small
How much VRAM does Moonshine Streaming Small need?
About 0.8 GB at F16 with an 8K context window. That is the 0.3 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 Moonshine Streaming Small on an 8 GB GPU?
Yes, with F32: the 0.6 GB file needs about 1.1 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Moonshine Streaming Small GGUF?
It depends on the quant: Q8_0 is 0.2 GB, F16 is 0.3 GB, F32 is 0.6 GB.
Where can I download Moonshine Streaming Small GGUF?
The files are on Hugging Face in the handy-computer/moonshine-streaming-small-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 Moonshine Streaming Small use?
Hugging Face lists the license as mit. 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 Moonshine Streaming Small on 4 GB VRAM?
Yes, the Q8_0 quant uses roughly 0.7 GB of VRAM at 8 k context, which fits within a 4 GB budget.
Does the model support multiple languages?
No, it is trained and evaluated only on English speech.
What audio format is required?
The model expects 16 kHz mono WAV files; other formats must be converted first.