Skip to content

Fix beam search output_log_prob index error #732

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
cpm0722 wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Fix beam search output_log_prob index error #732

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 3 additions & 1 deletion src/fastertransformer/layers/beam_search_layers/OnlineBeamSearchLayer.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -115,6 +115,7 @@ void OnlineBeamSearchLayer<T>::invokeSoftMax(TensorMap* output_tensors, TensorMa

const int batch_size = output_tensors->at("output_ids").shape[1];
const int beam_width = output_tensors->at("output_ids").shape[2];
const int max_input_length = input_tensors->at("max_input_length").getVal<int>();
const int step = input_tensors->at("step").getVal<int>();
const int ite = input_tensors->at("ite").getVal<int>();
const int local_batch_size = input_tensors->at("logits").shape[0];
Expand All @@ -125,6 +126,7 @@ void OnlineBeamSearchLayer<T>::invokeSoftMax(TensorMap* output_tensors, TensorMa
input_tensors->isExist("len_penalty") ? input_tensors->at("len_penalty").getVal<float>() : 0.0f;

const int id_offset = step * batch_size * beam_width + local_batch_size * ite * beam_width;
const int gen_offset = (step - max_input_length) * batch_size * beam_width + local_batch_size * ite * beam_width;

BeamHypotheses beam_hyps;
if (output_tensors->isExist("beam_hyps")) {
Expand All @@ -147,7 +149,7 @@ void OnlineBeamSearchLayer<T>::invokeSoftMax(TensorMap* output_tensors, TensorMa
output_tensors->at("finished").getPtr<bool>(),
output_tensors->at("sequence_length").getPtr<int>(),
output_tensors->at("cum_log_probs").getPtr<float>(),
output_tensors->getPtrWithOffset<float>("output_log_probs", id_offset, nullptr),
output_tensors->getPtrWithOffset<float>("output_log_probs", gen_offset, nullptr),
output_tensors->at("output_ids").getPtrWithOffset<int>(id_offset),
topk_softmax_workspace_,
topk_softmax_workspace_size_,
Expand Down