How much VRAM does SmolVLM2-2.2B-Instruct need?

Q4_K_M is exactly 1,112,602,656 bytes (1.04 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

HuggingFaceTB · text

SmolVLM2-2.2B-Instruct

HuggingFaceTB/SmolVLM2-2.2B-Instruct

SmolVLM2-2.2B-Instruct at Q4_K_M is exactly 1,112,602,656 bytes (1.04 GiB / 1.11 GB) — an effective 3.962 bits per weight, not the nominal 4.

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

Parameters

2.2B

Architecture

llama

24 layers

Context

8,192

native (config.json)

License

apache-2.0

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
I1-IQ1_S	0.42 GiB	445,612,768	1.587	—	mradermacher
I1-IQ1_M	0.44 GiB	477,463,264	1.700	—	mradermacher
I1-IQ2_XXS	0.49 GiB	530,547,424	1.889	—	mradermacher
I1-IQ2_XS	0.54 GiB	576,160,480	2.051	—	mradermacher
I1-IQ2_S	0.57 GiB	615,901,920	2.193	—	mradermacher
I1-IQ2_M	0.61 GiB	658,369,248	2.344	—	mradermacher
I1-Q2_K_S	0.61 GiB	658,377,440	2.344	—	mradermacher
Q2_K	0.66 GiB	707,921,984	2.521	—	second-state
I1-Q2_K	0.66 GiB	707,922,656	2.521	—	mradermacher
I1-IQ3_XXS	0.67 GiB	723,643,104	2.577	—	mradermacher
I1-IQ3_XS	0.73 GiB	782,660,320	2.787	—	mradermacher
Q3_K_S	0.76 GiB	820,408,384	2.921	—	second-state
I1-Q3_K_S	0.76 GiB	820,409,056	2.921	—	mradermacher
I1-IQ3_S	0.76 GiB	820,409,056	2.921	—	mradermacher
I1-IQ3_M	0.80 GiB	853,832,416	3.040	—	mradermacher
Q3_K_M	0.84 GiB	903,770,176	3.218	—	second-state
I1-Q3_K_M	0.84 GiB	903,770,848	3.218	—	mradermacher
Q3_K_L	0.91 GiB	976,121,920	3.476	—	second-state
I1-Q3_K_L	0.91 GiB	976,122,592	3.476	—	mradermacher
I1-IQ4_XS	0.93 GiB	994,237,152	3.540	—	mradermacher
Q4_0	0.98 GiB	1,047,722,048	3.731	—	second-state
I1-IQ4_NL	0.98 GiB	1,047,722,720	3.731	—	mradermacher
I1-Q4_0	0.98 GiB	1,050,868,448	3.742	—	mradermacher
Q4_K_S	0.98 GiB	1,056,110,656	3.760	—	second-state
I1-Q4_K_S	0.98 GiB	1,056,111,328	3.760	—	mradermacher
Q4_K_M	1.04 GiB	1,112,602,656	3.962	—	ggml-org
Q4_K_M	1.04 GiB	1,112,602,688	3.962	—	second-state
I1-Q4_K_M	1.04 GiB	1,112,603,360	3.962	—	mradermacher
I1-Q4_1	1.08 GiB	1,154,693,856	4.112	—	mradermacher
Q5_0	1.18 GiB	1,261,664,320	4.492	—	second-state
Q5_K_S	1.18 GiB	1,261,664,320	4.492	—	second-state
I1-Q5_K_S	1.18 GiB	1,261,664,992	4.492	—	mradermacher
Q5_K_M	1.21 GiB	1,295,087,680	4.611	—	second-state
I1-Q5_K_M	1.21 GiB	1,295,088,352	4.611	—	mradermacher
Q6_K	1.39 GiB	1,488,977,984	5.302	—	second-state
I1-Q6_K	1.39 GiB	1,488,978,656	5.302	—	mradermacher
Q8_0	1.80 GiB	1,927,933,984	6.865	—	ggml-org
Q8_0	1.80 GiB	1,927,934,016	6.865	—	second-state
F16	3.38 GiB	3,627,118,624	12.915	—	ggml-org
F16	3.38 GiB	3,627,118,656	12.915	—	second-state

Compare with

same modality, comparable size

vs Qwen3-4B vs Llama-3.2-1B-Instruct vs ced-base vs gemma-3-1b-it vs Qwen3-1.7B vs embeddinggemma-300m 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 1.18 GiB. The real file is 1.04 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

2048

Vocab

49,280

Sliding window

none

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does SmolVLM2-2.2B-Instruct need?: Q4_K_M is exactly 1,112,602,656 bytes (1.04 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
Which quantization of SmolVLM2-2.2B-Instruct 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.