How much VRAM does LFM2-VL-1.6B need?

Q4_K_M is exactly 730,894,016 bytes (0.68 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is LFM2-VL-1.6B's KV cache?

0.38 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.

LiquidAI · vision language

LFM2-VL-1.6B

LiquidAI/LFM2-VL-1.6B

LFM2-VL-1.6B at Q4_K_M is exactly 730,894,016 bytes (0.68 GiB / 0.73 GB) — an effective 3.689 bits per weight, not the nominal 4. Its KV cache at 32K is 0.38 GiB.

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

Parameters

1.6B

Architecture

lfm2

16 layers

Context

128,000

native (config.json)

License

other

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
IQ2_M	0.40 GiB	434,130,624	2.192	—	bartowski
Q2_K	0.45 GiB	483,397,312	2.440	—	bartowski
IQ3_XXS	0.46 GiB	490,983,104	2.478	—	bartowski
Q2_K_L	0.48 GiB	515,903,168	2.604	—	bartowski
IQ3_XS	0.50 GiB	537,808,576	2.715	—	bartowski
Q3_K_S	0.52 GiB	558,157,504	2.817	—	bartowski
IQ3_M	0.53 GiB	566,791,872	2.861	—	bartowski
Q3_K_M	0.56 GiB	600,346,304	3.030	—	bartowski
Q3_K_L	0.59 GiB	635,473,600	3.208	—	bartowski
IQ4_XS	0.62 GiB	663,375,552	3.349	—	bartowski
Q4_0	0.65 GiB	695,750,048	3.512	—	LiquidAI
IQ4_NL	0.65 GiB	695,750,336	3.512	—	bartowski
Q4_0	0.65 GiB	697,847,488	3.523	—	bartowski
Q4_K_S	0.65 GiB	700,468,928	3.536	—	bartowski
Q4_K_M	0.68 GiB	730,894,016	3.689	—	bartowski
Q4_1	0.71 GiB	760,499,904	3.839	—	bartowski
Q4_K_L	0.71 GiB	763,399,872	3.854	—	bartowski
Q5_K_S	0.77 GiB	825,249,472	4.166	—	bartowski
Q5_K_M	0.79 GiB	843,353,792	4.257	—	bartowski
Q5_K_L	0.82 GiB	875,859,648	4.421	—	bartowski
Q6_K	0.90 GiB	962,842,304	4.860	—	bartowski
Q6_K_L	0.93 GiB	995,348,160	5.024	—	bartowski
Q8_0	1.16 GiB	1,246,252,448	6.291	—	LiquidAI
Q8_0	1.16 GiB	1,246,252,736	6.291	—	bartowski
F16	2.18 GiB	2,343,325,088	11.829	—	LiquidAI
BF16	2.18 GiB	2,343,325,088	11.829	—	bartowski

KV cache by context

computed per layer

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	0.05 GiB	0.13 GiB	2.67×	6 / 0 / 10
8,192	0.09 GiB	0.25 GiB	2.67×	6 / 0 / 10
16,384	0.19 GiB	0.50 GiB	2.67×	6 / 0 / 10
32,768	0.38 GiB	1.00 GiB	2.67×	6 / 0 / 10
65,536	0.75 GiB	2.00 GiB	2.67×	6 / 0 / 10
131,072	1.50 GiB	4.00 GiB	2.67×	6 / 0 / 10

10 of 16 layers use linear attention, which keeps a fixed-size recurrent state instead of a per-token cache. Those layers do not grow with context at all — treating them as ordinary attention, as a flat formula does, overstates this model's cache by roughly 2.7× at long context.

Compare with

same modality, comparable size

vs Qwen3.5-0.8B vs Qwen3-VL-4B-Instruct vs Qwen3-VL-2B-Instruct vs Qwen3.5-2B vs LFM2.5-VL-1.6B vs MiniCPM-V-4.6 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 0.83 GiB. The real file is 0.68 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

65,536

Sliding window

none

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does LFM2-VL-1.6B need?: Q4_K_M is exactly 730,894,016 bytes (0.68 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is LFM2-VL-1.6B's KV cache?: 0.38 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 LFM2-VL-1.6B 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.