Instructions to use mlx-community/DeepSeek-V3-4bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use mlx-community/DeepSeek-V3-4bit with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] huggingface-cli download --local-dir DeepSeek-V3-4bit mlx-community/DeepSeek-V3-4bit
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
VRAM Requirements for Running the Model
#1
by wilfoderek - opened
How much VRAM is needed to run the model?
Is an H200 sufficient?
Thank you in advance!
Same question here, it seems the model params size is ~200GB, how much VRAM is needed? Thanks
@awni says it only has 37B parameters in memory. I'm not sure how that translates to GB. I'm gonna give it a try.
No that’s not quite right. It only needs to move 37B parameters from RAM to cache. To run this thing you need about 400GB of RAM