How much VRAM does gemma-4-31B-it-qat-q4_0-unquantized-assistant need?

Q4_K_M is exactly 353,485,920 bytes (0.33 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-qat-q4_0-unquantized-assistant's KV cache?

0.57 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-4-31B-it-qat-q4_0-unquantized-assistant

google/gemma-4-31B-it-qat-q4_0-unquantized-assistant

gemma-4-31B-it-qat-q4_0-unquantized-assistant at Q4_K_M is exactly 353,485,920 bytes (0.33 GiB / 0.35 GB) — an effective 6.023 bits per weight, not the nominal 4. Its KV cache at 32K is 0.57 GiB, not the 2.00 GiB a flat formula predicts.

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

Parameters

470M

Architecture

gemma4-assistant

4 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	0.29 GiB	309,704,704	5.277	—	mradermacher
Q3_K_S	0.30 GiB	322,457,600	5.494	—	mradermacher
Q3_K_M	0.31 GiB	332,484,608	5.665	—	mradermacher
Q3_K_L	0.32 GiB	341,921,792	5.826	—	mradermacher
IQ4_XS	0.32 GiB	342,878,208	5.842	—	mradermacher
Q4_K_S	0.33 GiB	349,161,472	5.949	—	mradermacher
Q4_K_M	0.33 GiB	353,485,920	6.023	—	Gaboo
Q4_K_M	0.33 GiB	357,812,224	6.097	—	mradermacher
Q5_K_M	0.35 GiB	376,521,824	6.415	—	Gaboo
Q6_K	0.37 GiB	400,997,472	6.832	—	Gaboo
Q8_0	0.48 GiB	514,704,480	8.770	—	Gaboo
BF16	0.89 GiB	954,860,640	16.270	—	Gaboo
BF16	0.89 GiB	954,860,896	16.270	—	RachidAR

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	0.13 GiB	0.25 GiB	1.88×	1 / 3 / 0
8,192	0.20 GiB	0.50 GiB	2.56×	1 / 3 / 0
16,384	0.32 GiB	1.00 GiB	3.12×	1 / 3 / 0
32,768	0.57 GiB	2.00 GiB	3.51×	1 / 3 / 0
65,536	1.07 GiB	4.00 GiB	3.74×	1 / 3 / 0
131,072	2.07 GiB	8.00 GiB	3.86×	1 / 3 / 0

3 of 4 layers cache only a 1,024-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 Llama-3.2-1B-Instruct vs ced-base vs gemma-3-1b-it vs Qwen3-1.7B vs embeddinggemma-300m vs Qwen3-0.6B 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 0.25 GiB. The real file is 0.33 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 2.00 GiB at 32K context where the real figure is 0.57 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

1024

Vocab

262,144

Sliding window

1024

SWA period

—

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does gemma-4-31B-it-qat-q4_0-unquantized-assistant need?: Q4_K_M is exactly 353,485,920 bytes (0.33 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-qat-q4_0-unquantized-assistant's KV cache?: 0.57 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-qat-q4_0-unquantized-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.