AI Engineering ／ Newtype
newtype Model Evaluation, Fine-Tuning, and Hardware Evolution
Huang Yihe 2024-03 → 2026-05 Full evolution of model evaluation/fine-tuning/hardware real-world testing. 505 posts, covering the evolution of GPT/Claude/Gemini/Grok/DeepSeek/Qwen/Llama/Kimi/GLM/Minimax, Mac cluster benchmarks, Unsloth GRPO fine-tuning, Scaling Law debates, output length bottlenecks, TPU vs GPU, the strategy of general base models plus upper-layer vertical fine-tuning, model routing, and model committees.