Google · First MLX build 2 Apr 2026 · In Core AI since 23 Oct 2026
gemma-4-E2B-it
MLXCore AI
TextVision apache-2.0 Trending
A 5.1B parameter vision and text model from Google, with 10 MLX builds, from 4-bit to bf16 and Apple Core AI export recipes for iOS.
- Parameters
- 5.1B From Hugging Face
- Quantisation
- 4-bit gemma-4-e2b-it-4bit
- Memory needed
- 6 GB Estimate, unified memory
- Download
- 3.6 GB Weights, estimated
- Downloads
- 25K This build, last 30 days
- Task
- Vision and text Text, Vision
- MLX builds
- 10 On mlx-community
- Devices
- Mac Apple silicon
- Parameters
- 5.1B From Hugging Face
- Compression
- None Apple’s preset
- Memory needed
- — Parameter count not published
- Context
- 128K tokens Fixed at export
- Precision
- FP16 Compute precision
- Platforms
- iOS iPhone
- Export presets
- 1 In Apple’s registry
- Listed by Apple
- 23 Oct 2026 First registry commit
Memory
Will it fit?Needs about 6 GB at 4-bit
8 GB
16 GB
24 GB
32 GB
64 GB
128 GB
Fits with headroomTight, close other appsWon’t fit
Estimated from the size of the weights, plus a fifth for context and working memory and 2 GB for macOS and your apps. Long contexts need more.
Run it
Terminal · mlx-vlm$ pip install -U mlx-vlm
$ python -m mlx_vlm.generate --model mlx-community/gemma-4-e2b-it-4bit --image photo.jpg --prompt "Describe this image"Memory
There is no memory estimate: the parameter count for this model is not published on Hugging Face.
Export it
Terminal · uv$ git clone https://github.com/apple/coreai-models
$ cd coreai-models
$ uv run coreai.llm.export gemma4-e2b-it --platform iOS
Apple publishes the recipe, not a ready-made package: these commands convert the original weights to a Core AI .aimodel on your Mac. See the repository for requirements.
MLX builds
10 on mlx-community| gemma-4-e2b-it-OptiQ-4bit | 4-bit | 2.6 GB | 5 GB | 506 |
|---|---|---|---|---|
| Gemma4-E2B-IT-Text-int4 | Unstated | 2.6 GB | 5 GB | 481 |
| gemma-4-e2b-it-4bit | 4-bit | 3.6 GB | 6 GB | 25K |
| gemma-4-e2b-it-mxfp4 | MXFP4 | 3.6 GB | 6 GB | 200 |
| gemma-4-e2b-it-nvfp4 | 4-bit | 3.6 GB | 6 GB | 131 |
| gemma-4-e2b-it-5bit | 5-bit | 4.1 GB | 7 GB | 150 |
| gemma-4-e2b-it-6bit | 6-bit | 4.7 GB | 8 GB | 285 |
| gemma-4-e2b-it-8bit | 8-bit | 5.9 GB | 9 GB | 640 |
| gemma-4-e2b-it-mxfp8 | MXFP8 | 5.9 GB | 9 GB | 133 |
| gemma-4-e2b-it-bf16 | bf16 | 10.2 GB | 14 GB | 1.5K |
Sizes are estimated from Hugging Face’s parameter counts and the quantisation. Downloads are for the last 30 days.
Core AI export presets
From Apple’s registry| Preset | Platform | Compression | Precision | Context |
|---|---|---|---|---|
| gemma4-e2b-it Exported via models/gemma4/export.py (standalone recipe -- doesn't fit coreai.llm.export's single-graph pipeline: a per-context blocked ladder of statically-shaped programs, externalized Per-Layer Embeddings, and precomputed dual-RoPE rows). Pass --include-vision for image input. |
iOS | None | FP16 | 128K |