Skip to content

Support dynamic decoder layers in canary model runtime #3268

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

Merged
csukuangfj merged 2 commits into from
Mar 9, 2026
Merged

Support dynamic decoder layers in canary model runtime #3268

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
71e0750
support variable decoder layers in canary model
mm65x Mar 7, 2026
15576e0
use existing macro for metadata reading
mm65x Mar 8, 2026
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
2 changes: 2 additions & 0 deletions sherpa-onnx/csrc/offline-canary-model-meta-data.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,6 +14,8 @@ struct OfflineCanaryModelMetaData {
int32_t vocab_size;
int32_t subsampling_factor = 8;
int32_t feat_dim = 120;
int32_t num_decoder_layers = 6;
int32_t decoder_hidden_size = 1024;
std::string normalize_type;
std::unordered_map<std::string, int32_t> lang2id;
};
Expand Down
13 changes: 10 additions & 3 deletions sherpa-onnx/csrc/offline-canary-model.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -117,11 +117,13 @@ class OfflineCanaryModel::Impl {
}

std::vector<Ort::Value> GetInitialDecoderStates() {
std::array<int64_t, 3> shape{1, 0, 1024};
int32_t num_layers = meta_.num_decoder_layers;
int64_t hidden_size = meta_.decoder_hidden_size;
std::array<int64_t, 3> shape{1, 0, hidden_size};

std::vector<Ort::Value> ans;
ans.reserve(6);
for (int32_t i = 0; i < 6; ++i) {
ans.reserve(num_layers);
for (int32_t i = 0; i < num_layers; ++i) {
Ort::Value state = Ort::Value::CreateTensor<float>(
Allocator(), shape.data(), shape.size());

Expand Down Expand Up @@ -178,6 +180,11 @@ class OfflineCanaryModel::Impl {
"normalize_type");
SHERPA_ONNX_READ_META_DATA(meta_.subsampling_factor, "subsampling_factor");
SHERPA_ONNX_READ_META_DATA(meta_.feat_dim, "feat_dim");

SHERPA_ONNX_READ_META_DATA_WITH_DEFAULT(meta_.num_decoder_layers,
"num_decoder_layers", 6);
SHERPA_ONNX_READ_META_DATA_WITH_DEFAULT(meta_.decoder_hidden_size,
"decoder_hidden_size", 1024);
}

void InitDecoder(void *model_data, size_t model_data_length) {
Expand Down