Irfanuruchi/Phi-4-mini-instruct-MLX-4bit

Phi-4-mini-instruct (MLX 4-bit)

This is a 4-bit MLX quantized version of microsoft/Phi-4-mini-instruct, optimized for Apple Silicon and local / on-device inference.

Benchmark Environment

• Device: MacBook Pro (M3 Pro)
• Runtime: MLX
• Precision: 4-bit (~4.5 bits per weight)

Performance (Measured)

• Disk size: ~2.0 GB
• Peak memory: ~2.24 GB
• Generation speed: ~56 tokens/sec

Benchmarks were collected on macOS (M3 Pro).
iPhone / iPad performance will vary depending on hardware and memory.

Usage

mlx_lm.generate \
  --model Irfanuruchi/Phi-4-mini-instruct-MLX-4bit \
  --prompt "Give me 5 short offline assistant tips." \
  --max-tokens 120

License

Original model license applies. See microsoft/Phi-4-mini-instruct.