How much VRAM does LFM2.5-Embedding-350M need?

Q4_K_M is exactly 229,311,232 bytes (0.21 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

LiquidAI · embedding

LFM2.5-Embedding-350M

LiquidAI/LFM2.5-Embedding-350M

LFM2.5-Embedding-350M at Q4_K_M is exactly 229,311,232 bytes (0.21 GiB / 0.23 GB) — an effective 5.175 bits per weight, not the nominal 4.

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

Parameters

354M

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	0.15 GiB	160,597,184	3.624	—	mradermacher
Q3_K_S	0.17 GiB	181,150,912	4.088	—	mradermacher
Q3_K_M	0.18 GiB	193,152,192	4.359	—	mradermacher
Q3_K_L	0.19 GiB	203,048,128	4.582	—	mradermacher
IQ4_XS	0.20 GiB	211,805,376	4.780	—	mradermacher
Q4_0	0.20 GiB	219,308,800	4.949	—	LiquidAI
Q4_K_S	0.21 GiB	220,751,040	4.982	—	mradermacher
IQ4_XS	0.21 GiB	226,168,704	5.104	—	cstr
Q4_K_M	0.21 GiB	229,311,232	5.175	—	LiquidAI
Q4_K_M	0.21 GiB	229,311,680	5.175	—	mradermacher
Q4_K	0.22 GiB	235,147,136	5.307	—	cstr
Q5_K_S	0.24 GiB	255,222,976	5.760	—	mradermacher
Q5_K_M	0.24 GiB	260,375,296	5.876	—	LiquidAI
Q5_K_M	0.24 GiB	260,375,744	5.876	—	mradermacher
Q6_K	0.27 GiB	293,380,864	6.621	—	LiquidAI
Q6_K	0.27 GiB	293,381,312	6.621	—	mradermacher
Q8_0	0.35 GiB	378,802,048	8.549	—	cstr
Q8_0	0.35 GiB	379,216,640	8.558	—	LiquidAI
Q8_0	0.35 GiB	379,217,088	8.558	—	mradermacher
F16	0.66 GiB	711,069,504	16.047	—	cstr
BF16	0.66 GiB	711,484,160	16.057	—	LiquidAI
F16	0.66 GiB	711,484,160	16.057	—	LiquidAI
F16	0.66 GiB	711,484,608	16.057	—	mradermacher

No KV cache

architectural, not a gap in our data

This architecture allocates no KV cache. Encoder and embedding models process their input in one pass rather than generating token by token, so there is nothing to carry forward between steps and memory does not grow with context. Its footprint is the weights plus a working buffer, and that is the whole story.

Compare with

same modality, comparable size

vs jina-embeddings-v5-text-small vs embeddinggemma-300m-qat-q8_0-unquantized vs nomic-embed-text-v1.5 vs jina-embeddings-v5-text-nano vs all-MiniLM-L6-v2 vs mxbai-embed-xsmall-v1 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 0.19 GiB. The real file is 0.21 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

1024

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-Embedding-350M need?: Q4_K_M is exactly 229,311,232 bytes (0.21 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
Which quantization of LFM2.5-Embedding-350M 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.