maxmustermann/DeepLearningExamples
1
0
Fork 0
Code Issues Pull requests Projects Releases Wiki Activity

Fix: Fix the bugs of allocating workspace (#746,#747)

Browse source
This commit is contained in:
byshiue 2020-11-09 10:24:56 +08:00 committed by GitHub
parent 0b34777a51
commit a095658e44
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 2 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
FasterTransformer/v3.0/fastertransformer/decoding_beamsearch.h
View file

@ -114,8 +114,8 @@ public:
int finished_buf_size = args_.batch_size_ * args_.beam_width_; //type bool
int finished_count_size = (int)(ceil(1 / 32.)) * 32; // type int
int topk_ids_buf_size = args_.batch_size_ * args_.beam_width_ * (ceil)((args_.beam_width_ * args_.vocab_size_ * 1.0) / 1024.0); // type int
int topk_val_buf_size = args_.batch_size_ * args_.beam_width_ * args_.beam_width_; // type float
int topk_ids_buf_size = args_.batch_size_ * args_.beam_width_ * (ceil)((args_.beam_width_ * args_.vocab_size_ * 1.0) / 1024.0) * 8; // type int
int topk_val_buf_size = args_.batch_size_ * args_.beam_width_ * args_.beam_width_ * 8; // type float
int storage_size_per_beam = 2 * args_.beam_width_ + SMALL_TOP_K_SOFTMAX_MAX_VOC_PARTS * (2 * MAX_K + 2);
args_.temp_storage_size_ = args_.batch_size_ * args_.beam_width_ * storage_size_per_beam; // type float