Closed yzh119 closed 2 months ago
Add a new API ParallelTopPSamplingFromProb, which enables sampling from the same distribution multiple times, and allowing user to specify batch-specific top_p.
ParallelTopPSamplingFromProb
top_p
cc @MasterJH5574
Add a new API
ParallelTopPSamplingFromProb
, which enables sampling from the same distribution multiple times, and allowing user to specify batch-specifictop_p
.cc @MasterJH5574