How much VRAM does gemma-4-E4B-it-assistant need?

Q4_K_M is exactly 77,231,200 bytes (0.07 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is gemma-4-E4B-it-assistant's KV cache?

0.07 GiB at 32K context with an f16 cache, computed per layer from the model's architecture — most of its layers use sliding-window attention, so a flat formula overstates this substantially. Quantizing the cache to q8_0 roughly halves it.

google · text

gemma-4-E4B-it-assistant

google/gemma-4-E4B-it-assistant

gemma-4-E4B-it-assistant at Q4_K_M is exactly 77,231,200 bytes (0.07 GiB / 0.08 GB) — an effective 7.843 bits per weight, not the nominal 4. Its KV cache at 32K is 0.07 GiB, not the 0.25 GiB a flat formula predicts.

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

Parameters

79M

Architecture

gemma4-assistant

4 layers

Context

131,072

native (config.json)

License

apache-2.0

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
Q4_0	0.07 GiB	76,960,672	7.815	—	matrixportalx
Q4_K_S	0.07 GiB	76,960,864	7.815	—	AtomicChat
Q4_K_M	0.07 GiB	77,231,200	7.843	—	AtomicChat
Q5_K_M	0.07 GiB	78,460,000	7.968	—	AtomicChat
Q6_K	0.07 GiB	79,765,408	8.100	—	matrixportalx
Q8_0	0.09 GiB	98,653,088	10.018	—	matrixportalx
Q8_0	0.09 GiB	98,653,280	10.018	—	AtomicChat
F16	0.16 GiB	171,766,688	17.443	—	matrixportalx
F16	0.16 GiB	171,766,880	17.443	—	AtomicChat

KV cache by context

computed per layer — this model uses sliding-window attention

Context	KV cache (f16)●	Flat formula	Overstated by	Full / windowed / recurrent
4,096	0.01 GiB	0.03 GiB	2.29×	1 / 3 / 0
8,192	0.02 GiB	0.06 GiB	2.91×	1 / 3 / 0
16,384	0.04 GiB	0.13 GiB	3.37×	1 / 3 / 0
32,768	0.07 GiB	0.25 GiB	3.66×	1 / 3 / 0
65,536	0.13 GiB	0.50 GiB	3.82×	1 / 3 / 0
131,072	0.26 GiB	1.00 GiB	3.91×	1 / 3 / 0

3 of 4 layers cache only a 512-token window rather than the full context, on a period of . Figures assume the default configuration; --swa-full disables the saving entirely.

Compare with

same modality, comparable size

vs ced-base vs SmolLM2-135M-Instruct vs gemma-3-270m-it vs tinygemma3_cifar vs LFM2.5-230M vs TinyMistral-248M-v2-Instruct 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 0.04 GiB. The real file is 0.07 GiB, because a quantization is a mixture and some tensors are always kept at higher precision. The larger discrepancy is the cache: a flat formula gives 0.25 GiB at 32K context where the real figure is 0.07 GiB, because most of this model's layers cache a fixed window rather than the whole context.

Architecture

from config.json

Layers

Attention heads

KV heads

Head dim

256

Hidden size

256

Vocab

262,144

Sliding window

512

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does gemma-4-E4B-it-assistant need?: Q4_K_M is exactly 77,231,200 bytes (0.07 GiB) in weights. Add the KV cache, which depends on your context length, plus roughly half a gigabyte of runtime overhead.
How large is gemma-4-E4B-it-assistant's KV cache?: 0.07 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 gemma-4-E4B-it-assistant 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.