Summary
AntLingAGI and Alibaba reported 0.63-second weight loading for the Ling-2.6-1T FP8 model, up to 780 times faster than disk loading. The end-to-end engine restart fell to about 32 seconds by avoiding repeated disk reads.
Why it matters
Fast recovery reduces the operational penalty of serving very large models and improves utilisation of expensive inference hardware.
Related coverage
- GLM-5.3 doubles its reported KernelBench-Mega speedup
- UBS sees $4.1 trillion of hyperscaler AI infrastructure spending through 2028
- Microsoft receives first production Vera Rubin systems as Nvidia ramps the platform