Google’s latest Gemma, small yet strong for chat and generation
500K+
GGUF version by Unsloth
Gemma is a versatile AI model family designed for tasks like question answering, summarization, and reasoning. With open weights and responsible commercial use, it supports image-text input, a 128K token context, and over 140 languages.
Gemma 3 4B model can be used for:
| Attribute | Details |
|---|---|
| Provider | Google DeepMind |
| Architecture | Gemma3 |
| Cutoff date | - |
| Languages | 140 languages |
| Tool calling | ❌ |
| Input modalities | Text, Image |
| Output modalities | Text, Code |
| License | Gemma Terms |
| Model variant | Parameters | Quantization | Context window | VRAM¹ | Size |
|---|---|---|---|---|---|
ai/gemma3:4Bai/gemma3:4B-Q4_K_Mai/gemma3:latest | 4B | MOSTLY_Q4_K_M | 131K tokens | 3.83 GiB | 2.31 GB |
ai/gemma3:270M-F16 | 270M | MOSTLY_F16 | 33K tokens | 1.59 GiB | 511.46 MB |
ai/gemma3:270M-UD-IQ2_XXS | 270M | MOSTLY_IQ2_XXS | 33K tokens | 1.26 GiB | 165.54 MB |
ai/gemma3:270M-UD-Q4_K_XL | 270M | MOSTLY_Q4_K_M | 33K tokens | 1.33 GiB | 235.95 MB |
ai/gemma3:4B-F16 | 4B | MOSTLY_BF16 | 131K tokens | 8.75 GiB | 7.23 GB |
¹: VRAM estimated based on model characteristics.
latest→4B
First, pull the model:
docker model pull ai/gemma3
Then run the model:
docker model run ai/gemma3
For more information on Docker Model Runner, explore the documentation.
| Category | Benchmark | Value |
|---|---|---|
| General | MMLU | 59.6 |
| GSM8K | 38.4 | |
| ARC-Challenge | 56.2 | |
| BIG-Bench Hard | 50.9 | |
| DROP | 60.1 | |
| STEM & Code | MATH | 24.2 |
| MBPP | 46.0 | |
| HumanEval | 36.0 | |
| Multilingual | MGSM | 34.7 |
| Global-MMLU-Lite | 57.0 | |
| XQuAD (all) | 68.0 | |
| Multimodal | VQAv2 | 63.9 |
| TextVQA | 58.9 | |
| DocVQA | 72.8 |
Content type
Model
Digest
sha256:a353a8898…
Size
3.1 GB
Last updated
6 months ago
docker model pull ai/gemma3Pulls:
12,220
Last week