v2.2 · 2026-06-12

VISION / MULTIMODAL MODEL — RUNTIME REQUIREMENTS

·approximateDownloadGb includes the Q4_K_M model plus a matching multimodal projection file where applicable.

·RAM and VRAM figures are estimates based on GGUF file size plus conservative runtime and KV-cache overhead, not official hardware requirements.

Gemma 3 4B Instruct

gemma-3-4b-it

MULTIMODAL

Gemma 3 4B Instruct is a 4B multimodal model profile for local AI planning. This page records provider, license, context, quantization, hardware-fit estimates, setup hints, source links, and caveats so readers do not choose a model by name alone.

PROVIDER

Google

FAMILY

Gemma 3

MODEL TYPE

multimodal

PARAMETERS

MODALITIES

text, image, video

ARCHITECTURE

decoder-only Transformer vision-language instruction model

CONTEXT WINDOW

131,072 tokens

TRAINING TOKENS

4.0T

RELEASE DATE

2025-03-10

LICENSE & LINKS

Gemma Terms of Uselicense text

MODEL CARD

ai.google.dev/gemma/docs/core/model_card_3

HUGGING FACE

google/gemma-3-4b-it

GGUF REPOSITORIES

bartowski/google_gemma-3-4b-it-GGUF community conversion

Community llama.cpp GGUF quantization of the official Google instruction-tuned checkpoint; image input requires a compatible mmproj file in llama.cpp-style runners.

SETUP HINTS

OLLAMA

ollama run gemma3:4b

LM STUDIO

Search for bartowski/google_gemma-3-4b-it-GGUF and verify image/mmproj support before enabling vision.

LLAMA.CPP

Use hf.co/bartowski/google_gemma-3-4b-it-GGUF:Q4_K_M plus the matching mmproj file for image input.

QUANTIZATION FILE SIZES (GGUF)

Q4_K_M

2.49 GB

Q5_K_M

2.83 GB

Q8_0

4.13 GB

Quantization note: Q4_K_M for default local use; Q5_K_M if the model and multimodal projection fit comfortably in RAM/VRAM.

RAM / VRAM ESTIMATES

MIN RAM

6 GB

COMFORTABLE RAM

12 GB

MIN VRAM

4 GB

COMFORTABLE VRAM

8 GB

These are conservative local-inference estimates, not official hardware requirements.

HARDWARE FIT

CPU only (no GPU)

yes

8 GB RAM

borderline

16 GB RAM

yes

32 GB RAM

yes

8 GB VRAM

yes

12 GB VRAM

yes

24 GB VRAM

yes

Apple Silicon (unified memory)

yes

Hardware fit values are conservative local-inference estimates based on GGUF size plus runtime and KV-cache overhead. Actual requirements depend on context length, quantization, runtime, GPU offload, and other running apps.

BEST FOR

·local multimodal chat

·image understanding

·structured output

·function-calling workflows

AVOID IF

·a text-only runner is required

·8GB RAM is the hard ceiling

·maximum answer quality is required

CAVEATS

·Video understanding is described by Google for Gemma 3 generally, but local GGUF/Ollama support should be treated as runner-dependent.

·GGUF files are community conversions, not official Google GGUF releases.