Higher acceptance length, slower prose: Ling’s n=1/2/3 MTP test on one Spark

The missing control is visible in sudoingX’s Ling-3.0-flash benchmark graphics. The earlier table leaves Ling’s no-speculation baseline as “not measured.” The later code/prose graphic fills it in: about 23 tok/s without the drafter, against 40.9 on code and 38.7 on prose with MTP n=1, at short

中文全文翻译正在补齐，可先查看下方英文原文。

The missing control is visible in sudoingX’s Ling-3.0-flash benchmark graphics. The earlier table leaves Ling’s no-speculation baseline as “not measured.” The later code/prose graphic fills it in: about 23 tok/s without the drafter, against 40.9 on code and 38.7 on prose with MTP n=1, at short