TLDR Tech / Practical
Profiling gpt-oss-120b locally on Apple M4 Max
The author describes running a profiling pipeline on the gpt-oss-120b model using an Apple M4 Max with llama.cpp, comparing its performance against Claude and other models. They observed that while local models sometimes terminate prematurely or have issues with client-side rendering, converting data retrieval and verification into deterministic scripts enabled the free local model to achieve competitive results with advanced models.
Read the full story