How much VRAM does gemma-3-4b-it-qat need?

Q4_0 is exactly 2,526,080,992 bytes (2.35 GiB) in weights. Add the KV cache, which depends on context length, plus roughly half a gigabyte of runtime overhead.

How large is gemma-3-4b-it-qat's KV cache?

0.79 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-3-4b-it-qat

google/gemma-3-4b-it-qat

gemma-3-4b-it-qat at Q4_0 is exactly 2,526,080,992 bytes (2.35 GiB / 2.53 GB) — an effective 5.208 bits per weight, not the nominal 4. Its KV cache at 32K is 0.79 GiB, not the 4.25 GiB a flat formula predicts.

From the file· summed from 1 file(s)From the file· KV from mirror (mirror:unsloth/gemma-3-4b-it-qat)

Parameters

3.9B

Architecture

gemma3

34 layers

Context

131,072

native (config.json)

License

—

Shipped quantizations

● exact bytes, summed from published files

Quant	Size●	Exact bytes●	Effective bpw●	Tensors●	Publisher
Q4_0	2.35 GiB	2,526,080,992	5.208	—	ggml-org

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	0.25 GiB	0.53 GiB	2.14×	5 / 29 / 0
8,192	0.33 GiB	1.06 GiB	3.26×	5 / 29 / 0
16,384	0.48 GiB	2.13 GiB	4.40×	5 / 29 / 0
32,768	0.79 GiB	4.25 GiB	5.35×	5 / 29 / 0
65,536	1.42 GiB	8.50 GiB	5.99×	5 / 29 / 0
131,072	2.67 GiB	17.00 GiB	6.37×	5 / 29 / 0

29 of 34 layers cache only a 1,024-token window rather than the full context, on a period of 6. Figures assume the default configuration; --swa-full disables the saving entirely.

Compare with

same modality, comparable size

vs gemma-4-E4B-it vs Qwen3-4B vs Qwen3-8B vs HyperCLOVAX-SEED-Text-Instruct-1.5B vs Llama-3.2-1B-Instruct vs llama-3-youko-8b All alternatives →Will it fit on my card? →

Why other calculators give a different number

A parameters × bits ÷ 8 estimate puts Q4_0 at roughly 2.03 GiB. The real file is 2.35 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 4.25 GiB at 32K context where the real figure is 0.79 GiB, because most of this model's layers cache a fixed window rather than the whole context.

Architecture

from mirror:unsloth/gemma-3-4b-it-qat

Layers

Attention heads

KV heads

Head dim

256

Hidden size

2560

Vocab

262,208

Sliding window

1024

SWA period

MLA

Experts

—

Experts per token

—

use_sliding_window

—

Questions people ask

How much VRAM does gemma-3-4b-it-qat need?: Q4_0 is exactly 2,526,080,992 bytes (2.35 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-3-4b-it-qat's KV cache?: 0.79 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-3-4b-it-qat 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.