DeepSeek · First MLX build 5 Jan 2025
DeepSeek-V3
MLX
A 671B parameter model from DeepSeek, with 3 MLX builds, from 3-bit to 4-bit.
- Parameters
- 671B From Hugging Face
- Quantisation
- 4-bit DeepSeek-V3-4bit
- Memory needed
- 425 GB Estimate, unified memory
- Download
- 378 GB Weights, estimated
- Downloads
- 324 This build, last 30 days
- Task
- —
- MLX builds
- 3 On mlx-community
- Devices
- Mac Apple silicon
Memory
Will it fit?Needs about 425 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 · Hugging Face CLI$ pip install -U huggingface_hub
$ hf download mlx-community/DeepSeek-V3-4bitMLX builds
3 on mlx-community| DeepSeek-V3-3bit | 3-bit | 294 GB | 331 GB | 78 |
|---|---|---|---|---|
| DeepSeek-V3-3bit-bf16 | 3-bit | 294 GB | 331 GB | 56 |
| DeepSeek-V3-4bit | 4-bit | 378 GB | 425 GB | 324 |
Sizes are estimated from Hugging Face’s parameter counts and the quantisation. Downloads are for the last 30 days.