How much VRAM does moonshine-streaming-tiny need?

Q4_K is exactly 32,052,896 bytes (0.03 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is moonshine-streaming-tiny's KV cache?

0.23 GiB at 32K context with an f16 cache, computed per layer from the model's architecture. Quantizing the cache to q8_0 roughly halves it.

UsefulSensors · audio asr

moonshine-streaming-tiny

UsefulSensors/moonshine-streaming-tiny

moonshine-streaming-tiny at Q4_K is exactly 32,052,896 bytes (0.03 GiB / 0.03 GB) — an effective 5.821 bits per weight, not the nominal 4. Its KV cache at 32K is 0.23 GiB.

From the file· summed from 1 file(s)From the file· KV per layer

Parameters

44M

Architecture

moonshine_streaming

6 layers

Context

4,096

native (config.json)

License

mit

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
Q4_K	0.03 GiB	32,052,896	5.821	—	cstr
Q8_0	0.05 GiB	50,462,816	9.164	—	handy-computer
F16	0.08 GiB	89,784,416	16.304	—	handy-computer
F32	0.17 GiB	177,817,696	32.290	—	handy-computer

KV cache by context

computed per layer

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	0.03 GiB	0.03 GiB	—	6 / 0 / 0
8,192	0.06 GiB	0.06 GiB	—	6 / 0 / 0
16,384	0.12 GiB	0.12 GiB	—	6 / 0 / 0
32,768	0.23 GiB	0.23 GiB	—	6 / 0 / 0
65,536	0.47 GiB	0.47 GiB	—	6 / 0 / 0
131,072	0.94 GiB	0.94 GiB	—	6 / 0 / 0

Compare with

same modality, comparable size

vs whisper-base vs whisper-tiny vs parakeet-tdt_ctc-110m vs whisper-base.en vs moonshine-base All alternatives →Will it fit on my card? →

Why other calculators give a different number

A parameters × bits ÷ 8 estimate puts Q4_K at roughly 0.02 GiB. The real file is 0.03 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.

Architecture

from config.json

Layers

Attention heads

KV heads

Head dim

Hidden size

320

Vocab

32,768

Sliding window

none

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does moonshine-streaming-tiny need?: Q4_K is exactly 32,052,896 bytes (0.03 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is moonshine-streaming-tiny's KV cache?: 0.23 GiB at 32K context with an f16 cache, computed per layer. Quantizing the cache to q8_0 roughly halves it, which is often the difference between a context length fitting and not.
Which quantization of moonshine-streaming-tiny should I use?: Q4_K_M is the usual default. Pick the largest quantization that fits your card at the context you actually need — the table above gives exact sizes for every one published.