GE

google_gemma-3-4b-it

GemmaGGUF by MungertUpdated 2025-09-24
Download on Hugging Face
Parameters3.88B
Context128K
Layers34
Architecturegemma3
Smallest quant1.9 GB
Quick answer

google_gemma-3-4b-it needs about 4.1 GB of memory at Q4_K_M (a 2.5 GB file) with an 8K context window, so it fits on a 6 GB GPU. This repository has 19 quants, from 1.9 GB (IQ3_XS) to 7.8 GB (BF16). Longer context needs more: at 32K tokens the KV cache alone adds about 4.6 GB.

Gemma‑3 4B Instruct model with experimental vision support

Gemma‑3 4B Instruct is a 3.88 billion‑parameter model that supports vision tasks via an experimental mmproj file. It is available in multiple GGUF quantizations and can be run with llama.cpp.

Good at

  • Instruct‑style text generation
  • Vision‑enabled prompts
  • Compact GGUF files

Limits

  • No official vision support in base model
  • Requires mmproj file for vision
  • License is gemma

Best for

  • Chat‑style Q&A
  • Image captioning with mmproj
  • Low‑latency inference on modest GPUs

Who it’s for. Developers building lightweight LLM applications who need both text and image understanding. The model is suitable for research and prototyping where experimental vision features are acceptable.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
IQ3_XS1.9 GB 3.5 GB
Fits Get file →
Q3_K_S1.9 GB 3.6 GB
Fits Get file →
Q3_K_M2.1 GB 3.7 GB
Fits Get file →
IQ4_XS2.3 GB 3.9 GB
Fits Get file →
IQ4_NL2.4 GB 4.0 GB
Fits Get file →
Q4_K_S2.4 GB 4.0 GB
Fits Get file →
Q4_K_M2.5 GB 4.1 GB
Fits Get file →
Q4_02.5 GB 4.2 GB
Fits Get file →
Q4_12.7 GB 4.4 GB
Fits Get file →
Q5_K_S2.8 GB 4.4 GB
Fits Get file →
Q5_K_M2.8 GB 4.5 GB
Fits Get file →
Q6_K_M3.2 GB 4.8 GB
Fits Get file →
F164.0 GB 5.6 GB
Fits Get file →
—4.1 GB 5.8 GB
Fits Get file →
BF164.8 GB 6.4 GB
Fits Get file →
F164.8 GB 6.4 GB
Fits Get file →
BF164.8 GB 6.4 GB
Fits Get file →
F164.8 GB 6.4 GB
Fits Get file →
BF167.8 GB 9.4 GB
Too big Get file →

For 8 GB VRAM, use Q4_K_M (2.49 GB file, ~4.1 GB memory at 8k context). For 12 GB VRAM, Q5_K_M (2.83 GB file, ~4.5 GB memory). For higher precision, BF16 (3.98 GB file, ~5.6 GB memory).

What GPU do you need for google_gemma-3-4b-it?

These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.

  • 4 GB VRAM: the best quant that fits is Q4_K_S (2.4 GB file, about 4.0 GB in use).
  • 6 GB VRAM: the best quant that fits is the main file (4.1 GB file, about 5.8 GB in use).
  • 8 GB VRAM: the best quant that fits is BF16 (4.8 GB file, about 6.4 GB in use).
  • 12 GB VRAM: the best quant that fits is BF16 (7.8 GB file, about 9.4 GB in use).

Which google_gemma-3-4b-it quant should you choose?

Q4_K_M is the usual starting point for google_gemma-3-4b-it: it balances file size and quality well. Here it is a 2.5 GB file.

If memory is tight, the smallest file is IQ3_XS at 1.9 GB, with the quality trade-offs that come with fewer bits. A reasonable choice when a 4-bit file will not fit, mostly on larger models. Quality loss is noticeable but often acceptable.

New to quant names? Read GGUF quantization explained.

Specifications

Architecture from the GGUF header

Architecturegemma3
Parameters3.88B
Layers34
Trained context131,072 tokens
KV cache (fp16)136 KB per token

Source

Quantized byMungert
FamilyGemma
Licensegemma
Downloads (HF)774
Last updated2025-09-24

Frequently asked questions about google_gemma-3-4b-it

How much VRAM does google_gemma-3-4b-it need?

About 4.1 GB at Q4_K_M with an 8K context window. That is the 2.5 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.

Can I run google_gemma-3-4b-it on an 8 GB GPU?

Yes, with BF16: the 4.8 GB file needs about 6.4 GB at 8K context. Larger quants in this repository do not fit in 8 GB.

What is the file size of google_gemma-3-4b-it GGUF?

It depends on the quant: IQ3_XS is 1.9 GB, Q3_K_S is 1.9 GB, Q3_K_M is 2.1 GB, IQ4_XS is 2.3 GB, IQ4_NL is 2.4 GB, Q4_K_S is 2.4 GB and 13 more.

What context length does google_gemma-3-4b-it support?

The GGUF header lists a trained context length of 131,072 tokens (128K). The KV cache grows with the context you use: about 1.1 GB at 8K and 4.6 GB at 32K.

Where can I download google_gemma-3-4b-it GGUF?

The files are on Hugging Face in the Mungert/gemma-3-4b-it-gguf repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.

What license does google_gemma-3-4b-it use?

Hugging Face lists the license as gemma. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.

Can I run Gemma‑3 4B with 8 GB VRAM?

Yes, using Q4_K_M or smaller quants keeps memory below 8 GB.

Does the model support vision out of the box?

The base model requires an mmproj file for vision; the GGUF file alone does not enable vision.

What is the context length?

The model was trained on 131072 tokens and supports that context length in inference.

Similar models

Browse all →

More from Mungert

All by Mungert →

More Gemma models

All Gemma →

Other small models

All small models →