AMD paper: FP4 training instability is not caused by insufficient randomness

AMD新论文颠覆认知：FP4训练不稳定，原因不是随机性不足

AMD and Penn State pretrained Llama 3.1-8B with MXFP4 on MI355X native FP4 hardware, achieving 9-10% end-to-end speedup over an FP8 baseline, while the paper identifies Wgrad quantization as the bottleneck that raises token overhead to 26-27% without deterministic Hadamard stabilization.

Why it matters: HKR-H/K/R all pass: a counterintuitive FP4 claim, concrete Llama 3.1-8B numbers, and a cost/hardware nerve. The topic is narrower training-infra research, so it stays in the 78-84 band.