Generate mixed‑precision quantization recipe for LLMs
Diagnose training run failures with causal replay bisect