How much VRAM does MiMo-VL-7B-RL need?

Q4_K_M is exactly 4,684,339,136 bytes (4.36 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is MiMo-VL-7B-RL's KV cache?

4.50 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.

XiaomiMiMo · vision language

MiMo-VL-7B-RL

XiaomiMiMo/MiMo-VL-7B-RL

MiMo-VL-7B-RL at Q4_K_M is exactly 4,684,339,136 bytes (4.36 GiB / 4.68 GB) — an effective 4.512 bits per weight, not the nominal 4. Its KV cache at 32K is 4.50 GiB.

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

Parameters

8.3B

Architecture

qwen2vl

36 layers

Context

128,000

native (config.json)

License

mit

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
I1-IQ1_S	1.86 GiB	2,002,169,504	1.928	—	mradermacher
I1-IQ1_M	1.98 GiB	2,129,768,096	2.051	—	mradermacher
UD-IQ1_S	1.99 GiB	2,141,665,216	2.063	—	unsloth
UD-IQ1_S	1.99 GiB	2,141,665,216	2.063	—	akpsahan
UD-IQ1_M	2.10 GiB	2,252,273,600	2.169	—	unsloth
UD-IQ1_M	2.10 GiB	2,252,273,600	2.169	—	akpsahan
I1-IQ2_XXS	2.18 GiB	2,342,432,416	2.256	—	mradermacher
UD-IQ2_XXS	2.28 GiB	2,445,686,720	2.356	—	akpsahan
UD-IQ2_XXS	2.28 GiB	2,445,686,720	2.356	—	unsloth
I1-IQ2_XS	2.36 GiB	2,531,438,240	2.438	—	mradermacher
I1-IQ2_S	2.51 GiB	2,697,789,088	2.598	—	mradermacher
I1-IQ2_M	2.67 GiB	2,867,920,544	2.762	—	mradermacher
I1-Q2_K_S	2.69 GiB	2,892,480,160	2.786	—	mradermacher
UD-IQ2_M	2.71 GiB	2,911,508,416	2.804	—	akpsahan
UD-IQ2_M	2.71 GiB	2,911,508,416	2.804	—	unsloth
Q2_K	2.87 GiB	3,076,405,184	2.963	—	akpsahan
Q2_K	2.87 GiB	3,076,405,184	2.963	—	unsloth
I1-Q2_K	2.87 GiB	3,076,406,944	2.963	—	mradermacher
I1-IQ3_XXS	2.94 GiB	3,152,543,392	3.036	—	mradermacher
UD-IQ3_XXS	2.97 GiB	3,189,282,752	3.072	—	akpsahan
UD-IQ3_XXS	2.97 GiB	3,189,282,752	3.072	—	unsloth
Q2_K_L	3.00 GiB	3,222,017,984	3.103	—	unsloth
Q2_K_L	3.00 GiB	3,222,017,984	3.103	—	akpsahan
I1-IQ3_XS	3.16 GiB	3,396,374,176	3.271	—	mradermacher
Q3_K_S	3.28 GiB	3,525,838,784	3.396	—	unsloth
Q3_K_S	3.28 GiB	3,525,838,784	3.396	—	akpsahan
I1-Q3_K_S	3.28 GiB	3,525,840,544	3.396	—	mradermacher
I1-IQ3_S	3.30 GiB	3,545,894,560	3.415	—	mradermacher
I1-IQ3_M	3.40 GiB	3,650,064,032	3.515	—	mradermacher
Q3_K_M	3.59 GiB	3,854,010,304	3.712	—	unsloth
Q3_K_M	3.59 GiB	3,854,010,304	3.712	—	akpsahan
I1-Q3_K_M	3.59 GiB	3,854,012,064	3.712	—	mradermacher
I1-Q3_K_L	3.85 GiB	4,138,962,592	3.986	—	mradermacher
I1-IQ4_XS	3.97 GiB	4,260,454,048	4.103	—	mradermacher
IQ4_XS	3.99 GiB	4,279,867,328	4.122	—	unsloth
IQ4_XS	3.99 GiB	4,279,867,328	4.122	—	akpsahan
Q4_0	4.16 GiB	4,466,907,072	4.302	—	akpsahan
Q4_0	4.16 GiB	4,466,907,072	4.302	—	unsloth
I1-Q4_0	4.16 GiB	4,466,908,832	4.302	—	mradermacher
IQ4_NL	4.17 GiB	4,474,509,248	4.310	—	unsloth

KV cache by context

computed per layer

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	0.56 GiB	0.56 GiB	—	36 / 0 / 0
8,192	1.13 GiB	1.13 GiB	—	36 / 0 / 0
16,384	2.25 GiB	2.25 GiB	—	36 / 0 / 0
32,768	4.50 GiB	4.50 GiB	—	36 / 0 / 0
65,536	9.00 GiB	9.00 GiB	—	36 / 0 / 0
131,072	18.00 GiB	18.00 GiB	—	36 / 0 / 0

Compare with

same modality, comparable size

vs Qwen3.5-9B vs gemma-4-12B-it vs Qwen3.5-4B vs Qwen3.5-0.8B vs gemma-4-E2B-it vs Qwythos-9B-v2 All alternatives →Will it fit on my card? →

Why other calculators give a different number

A parameters × bits ÷ 8 estimate puts Q4_K_M at roughly 4.35 GiB. The real file is 4.36 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

128

Hidden size

4096

Vocab

151,680

Sliding window

8192

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

false

This model declares a sliding window but sets use_sliding_window: false, so the window is not applied. Honouring the field without the flag understates KV for the whole family.

Questions people ask

How much VRAM does MiMo-VL-7B-RL need?: Q4_K_M is exactly 4,684,339,136 bytes (4.36 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is MiMo-VL-7B-RL's KV cache?: 4.50 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 MiMo-VL-7B-RL 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.