UsefulSensors · audio asr
moonshine-streaming-medium
UsefulSensors/moonshine-streaming-mediummoonshine-streaming-medium at Q8_0 is exactly 295,793,568 bytes (0.28 GiB / 0.30 GB) — an effective 8.899 bits per weight, not the nominal 8. Its KV cache at 32K is 1.09 GiB.
From the file· summed from 1 file(s)From the file· KV per layer
Parameters
266M
Architecture
moonshine_streaming
14 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 |
|---|---|---|---|---|---|
| Q8_0 | 0.28 GiB | 295,793,568 | 8.899 | — | handy-computer |
| F16 | 0.50 GiB | 533,781,408 | 16.060 | — | handy-computer |
| F32 | 0.99 GiB | 1,065,204,640 | 32.048 | — | handy-computer |
KV cache by context
computed per layer
| Context | KV cache (f16)● | Flat formula | Overstated by | Full / windowed / recurrent |
|---|---|---|---|---|
| 4,096 | 0.14 GiB | 0.14 GiB | — | 14 / 0 / 0 |
| 8,192 | 0.27 GiB | 0.27 GiB | — | 14 / 0 / 0 |
| 16,384 | 0.55 GiB | 0.55 GiB | — | 14 / 0 / 0 |
| 32,768 | 1.09 GiB | 1.09 GiB | — | 14 / 0 / 0 |
| 65,536 | 2.19 GiB | 2.19 GiB | — | 14 / 0 / 0 |
| 131,072 | 4.38 GiB | 4.38 GiB | — | 14 / 0 / 0 |
Measured
published by a third party, attributed below
| Metric | Value | What it means |
|---|---|---|
| rtf | 2681.0 | |
| RTFx | 2681.0 | Higher is better — audio seconds processed per second of compute. |
| Word error rate | 11.44% | Lower is better — the share of words transcribed incorrectly. |
| Word error rate | 8.14% | Lower is better — the share of words transcribed incorrectly. |
| Word error rate | 8.01% | Lower is better — the share of words transcribed incorrectly. |
| Word error rate | 9.75% | Lower is better — the share of words transcribed incorrectly. |
| Word error rate | 3.37% | Lower is better — the share of words transcribed incorrectly. |
| Word error rate | 8.93% | Lower is better — the share of words transcribed incorrectly. |
| Word error rate | 8.10% | Lower is better — the share of words transcribed incorrectly. |
| Word error rate | 2.26% | Lower is better — the share of words transcribed incorrectly. |
| Word error rate | 1.67% | Lower is better — the share of words transcribed incorrectly. |
| Word error rate | 4.60% | Lower is better — the share of words transcribed incorrectly. |
| Word error rate | 5.77% | Lower is better — the share of words transcribed incorrectly. |
Benchmarked· by open-asr-leaderboard-english-short-latest
RTFx measured by the Open ASR Leaderboard on a single datacenter GPU at a large batch size. It ranks models against each other; it says nothing about throughput on consumer hardware. We reproduce these figures with attribution; they are not ours and we have not verified the runs. Source: open-asr-leaderboard-english-short-latest.
Compare with
same modality, comparable size
Will it run on your card?
full quant x context sweep
Radeon RX 6500 XT 4GBGeForce RTX 3050 6GBGeForce RTX 5050 8GBGeForce RTX 3080 10GBGeForce RTX 2080 Ti 11GBGeForce RTX 5070 12GBGeForce RTX 5060 Ti 16GBApple M3 Pro 18GBGeForce RTX 3080 Ti 20GBGeForce RTX 5090 D V2 24GBGeForce RTX 5090 D 32GBApple M5 Max 36GBApple M5 Pro 48GBApple M5 Max 64GBApple M3 Ultra 96GBApple M5 Max 128GBApple M2 Ultra 192GBApple M3 Ultra 256GBApple M3 Ultra 512GB
Why other calculators give a different number
A parameters × bits ÷ 8 estimate puts Q8_0 at roughly 0.14 GiB. The real file is 0.28 GiB, because a quantization is a mixture and some tensors are always kept at higher precision.
Architecture
from config.json
Layers
14
Attention heads
10
KV heads
10
Head dim
64
Hidden size
640
Vocab
32,768
Sliding window
none
SWA period
—
MLA
no
Experts
—
Experts per token
—
use_sliding_window
—
Questions people ask
- How much VRAM does moonshine-streaming-medium need?
- Q8_0 is exactly 295,793,568 bytes (0.28 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-medium's KV cache?
- 1.09 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-medium 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.