How much VRAM does gemma-4-31B-it-DFlash need?

Q4_K_M is exactly 911,587,296 bytes (0.85 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is gemma-4-31B-it-DFlash's KV cache?

0.16 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.

z-lab · text

gemma-4-31B-it-DFlash

z-lab/gemma-4-31B-it-DFlash

gemma-4-31B-it-DFlash at Q4_K_M is exactly 911,587,296 bytes (0.85 GiB / 0.91 GB) — an effective 4.748 bits per weight, not the nominal 4. Its KV cache at 32K is 0.16 GiB, not the 0.63 GiB a flat formula predicts.

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

Parameters

1.5B

Architecture

dflash

5 layers

Context

262,144

native (config.json)

License

apache-2.0

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
Q2_K	0.54 GiB	578,185,184	3.011	—	Anbeeld
Q3_K_M	0.70 GiB	748,668,896	3.899	—	Anbeeld
Q4_K_M	0.85 GiB	911,587,296	4.748	—	Alittlehammmer
Q4_K_M	0.85 GiB	912,636,896	4.753	—	Anbeeld
Q5_K	1.01 GiB	1,087,748,064	5.665	—	Alittlehammmer
Q5_K_M	1.01 GiB	1,088,797,664	5.671	—	Anbeeld
Q6_K	1.19 GiB	1,274,918,880	6.640	—	Alittlehammmer
Q6_K	1.19 GiB	1,275,968,480	6.646	—	Anbeeld
Q8_0	1.53 GiB	1,646,895,072	8.578	—	Alittlehammmer
Q8_0	1.53 GiB	1,647,944,672	8.583	—	Anbeeld
BF16	2.87 GiB	3,086,802,912	16.077	—	Alittlehammmer
BF16	2.88 GiB	3,087,852,512	16.083	—	Anbeeld

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	0.05 GiB	0.08 GiB	1.43×	1 / 4 / 0
8,192	0.07 GiB	0.16 GiB	2.22×	1 / 4 / 0
16,384	0.10 GiB	0.31 GiB	3.08×	1 / 4 / 0
32,768	0.16 GiB	0.63 GiB	3.81×	1 / 4 / 0
65,536	0.29 GiB	1.25 GiB	4.32×	1 / 4 / 0
131,072	0.54 GiB	2.50 GiB	4.64×	1 / 4 / 0

4 of 5 layers cache only a 2,048-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 Qwen3-4B vs HyperCLOVAX-SEED-Text-Instruct-1.5B vs Llama-3.2-1B-Instruct vs ced-base vs gemma-3-1b-it vs Qwen3-1.7B 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 0.80 GiB. The real file is 0.85 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 0.63 GiB at 32K context where the real figure is 0.16 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

128

Hidden size

5376

Vocab

262,144

Sliding window

2048

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

true

Questions people ask

How much VRAM does gemma-4-31B-it-DFlash need?: Q4_K_M is exactly 911,587,296 bytes (0.85 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-4-31B-it-DFlash's KV cache?: 0.16 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-4-31B-it-DFlash 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.