From the source
Baidu ERNIE 4.5 models get a major inference speed boost using a new sparse attention mechanism called PLAS, available for -Paddle versions deployed with FastDeploy.
From the source
We’ve just rolled out a powerful update for the ERNIE-4.5-21B-A3B and ERNIE-4.5-300B-A47B models that speeds up inference for long-context tasks.
ernie.baidu.com