FasterTransformer

fix: change int of some kernels to int64_t to prevent overflow

Update beam_search_topk_kernels.cu

Merge branch 'main' of https://github.com/NVIDIA/FasterTransformer into main

fix: fix custom_ar_kernel threshold

fix: scale input bug inside AddBiasResidualLayerNorm kernel (#466)