Skip to content

DEV Community

Local Model Benchmarks Series' Articles

Back to Mihai Perdum's Series

Cover image for The best 4-bit format in MLX is the one its own converter sets up to lose

Sep 18

The best 4-bit format in MLX is the one its own converter sets up to lose

#aicoding

7 min read

Cover image for MLX memory ceiling on a 96 GB M3 Ultra

Sep 19

MLX memory ceiling on a 96 GB M3 Ultra

#aicoding #localai

15 min read

Cover image for llama.cpp vs MLX on Qwen3.6-27B: MTP is 1.04x here, not 1.85x

Sep 20

llama.cpp vs MLX on Qwen3.6-27B: MTP is 1.04x here, not 1.85x

#aicoding #localai

18 min read

Cover image for Ling 3.0 Flash won't load on a Mac Studio. Qwen3-Coder-Next does, at 73 tok/s.

Sep 20

Ling 3.0 Flash won't load on a Mac Studio. Qwen3-Coder-Next does, at 73 tok/s.

#aicoding #localai

18 min read

Cover image for MXFP8 vs Q8: 10x the weight error, 1% the perplexity

Sep 20

MXFP8 vs Q8: 10x the weight error, 1% the perplexity

#aicoding #localai

14 min read

Cover image for MLX vs GGUF on Apple Silicon: Benchmarking the Same Local Model Two Ways

Sep 20

MLX vs GGUF on Apple Silicon: Benchmarking the Same Local Model Two Ways

#ai #mlx #benchmark #localai

7 min read

Cover image for Qwen3-Coder-Next + MXFP8: The 128GB Local LLM That Runs Predictably

Sep 20

Qwen3-Coder-Next + MXFP8: The 128GB Local LLM That Runs Predictably

#aicoding #localai #llm #mlx

25 min read