How much VRAM does LFM2.5-VL-1.6B need?

Q4_K_M is exactly 730,895,968 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 LFM2.5-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

LFM2.5-VL-1.6B

LiquidAI/LFM2.5-VL-1.6B

LFM2.5-VL-1.6B at Q4_K_M is exactly 730,895,968 bytes (0.68 GiB / 0.73 GB) — an effective 3.662 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
Q2_K	0.45 GiB	483,399,264	2.422	—	unsloth
Q2_K_L	0.45 GiB	483,399,264	2.422	—	unsloth
Q3_K_S	0.52 GiB	558,159,456	2.797	—	unsloth
Q3_K_M	0.56 GiB	600,348,256	3.008	148	unsloth
Q4_0	0.65 GiB	695,752,288	3.486	148	unsloth
Q4_0	0.65 GiB	695,752,480	3.486	148	LiquidAI
Q4_K_S	0.65 GiB	700,470,880	3.510	—	unsloth
Q4_K_M	0.68 GiB	730,895,968	3.662	148	unsloth
Q4_K_M	0.68 GiB	730,896,256	3.662	148	LiquidAI
Q4_K_M	0.68 GiB	730,897,088	3.662	—	lmstudio-community
Q4_1	0.71 GiB	760,501,856	3.811	—	unsloth
Q5_K_S	0.77 GiB	825,251,424	4.135	—	unsloth
Q5_K_M	0.79 GiB	843,355,744	4.226	148	unsloth
Q5_K_M	0.79 GiB	843,356,032	4.226	148	LiquidAI
Q6_K	0.90 GiB	962,844,256	4.824	148	unsloth
Q6_K	0.90 GiB	962,844,544	4.824	148	LiquidAI
Q6_K	0.90 GiB	962,845,376	4.824	—	lmstudio-community
Q8_0	1.16 GiB	1,246,254,688	6.244	148	unsloth
Q8_0	1.16 GiB	1,246,254,880	6.244	148	LiquidAI
Q8_0	1.16 GiB	1,246,255,808	6.244	—	lmstudio-community
BF16	2.18 GiB	2,343,327,328	11.741	—	unsloth
BF16	2.18 GiB	2,343,327,520	11.741	—	LiquidAI
F16	2.18 GiB	2,343,327,520	11.741	148	LiquidAI

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 MiniCPM-V-4.6 vs OvisOCR2 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 0.84 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 LFM2.5-VL-1.6B need?: Q4_K_M is exactly 730,895,968 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 LFM2.5-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 LFM2.5-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.