MLX and MLX LM on DGX Spark

mlx_dgx_spark.md

To install mlx-lm

pip install mlx-lm[cuda13]

Run generation:

mlx_lm.generate \
  --model mlx-community/Qwen3-4B-Instruct-2507-mxfp8 \
  --prompt "Tell me a story about Einstein" \
  -m 1024 \
  --quantize-activations

LoRA fine-tune:

mlx_lm.lora \
  --model Qwen/Qwen3-4B-Instruct-2507 \
  --data mlx-community/WikiSQL \
  --train

Awesome, thanks! Struggled with this until I found your solution.