MLX Community · First MLX build 22 Dec 2024
deepseek-vl2
MLX
TextVision deepseek
A 27B parameter vision and text model, with 5 MLX builds, from 3-bit to bf16.
- Parameters
- 27B From Hugging Face
- Quantisation
- 8-bit deepseek-vl2-8bit
- Memory needed
- 35 GB Estimate, unified memory
- Download
- 29.3 GB Weights, estimated
- Downloads
- 195 This build, last 30 days
- Task
- Vision and text Text, Vision
- MLX builds
- 5 On mlx-community
- Devices
- Mac Apple silicon
Memory
Will it fit?Needs about 35 GB at 8-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/deepseek-vl2-8bit --image photo.jpg --prompt "Describe this image"MLX builds
5 on mlx-community| deepseek-vl2-3bit | 3-bit | 12.3 GB | 16 GB | 74 |
|---|---|---|---|---|
| deepseek-vl2-4bit | 4-bit | 15.7 GB | 20 GB | 141 |
| deepseek-vl2-6bit | 6-bit | 22.5 GB | 28 GB | 95 |
| deepseek-vl2-8bit | 8-bit | 29.3 GB | 35 GB | 195 |
| deepseek-vl2-bf16 | bf16 | 55.0 GB | 64 GB | 73 |
Sizes are estimated from Hugging Face’s parameter counts and the quantisation. Downloads are for the last 30 days.