Benchmarks and serving notes for Qwen3.5-27B fine-tune on Apple Silicon via MLX. Companion to dgx-spark-nvfp4-serving.
benchmarks quantization dora mlx fine-tuning apple-silicon mac-studio llama-cpp llm-inference speculative-decoding mlx-lm qwen3 qwen3-5 m3-ultra fastmtp
-
Updated
Apr 21, 2026 - Python