EngineX-Iluvatar/enginex-mr_series-sherpa-onnx
Archived
16
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

Support building GPU-capable sherpa-onnx on Linux aarch64. (#1500)

Browse Source 
Thanks to @Peakyxh for providing pre-built onnxruntime libraries 
with CUDA support for Linux aarch64.

Tested on Jetson nano b01
This commit is contained in:
Fangjun Kuang
2024-11-01 11:16:28 +08:00
committed by GitHub
parent a3c89aa0d8
commit 9ab89c33bc
41 changed files with 537 additions and 291 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
sherpa-onnx/csrc/online-conformer-transducer-model.cc
View File

@@ -163,8 +163,11 @@ std::vector<Ort::Value> OnlineConformerTransducerModel::StackStates(
conv_vec[i] = &states[i][1];
}
Ort::Value attn = Cat(allocator_, attn_vec, 2);
Ort::Value conv = Cat(allocator_, conv_vec, 2);
auto allocator =
const_cast<OnlineConformerTransducerModel *>(this)->allocator_;
Ort::Value attn = Cat(allocator, attn_vec, 2);
Ort::Value conv = Cat(allocator, conv_vec, 2);
std::vector<Ort::Value> ans;
ans.reserve(2);
@@ -183,8 +186,11 @@ OnlineConformerTransducerModel::UnStackStates(
std::vector<std::vector<Ort::Value>> ans(batch_size);
std::vector<Ort::Value> attn_vec = Unbind(allocator_, &states[0], 2);
std::vector<Ort::Value> conv_vec = Unbind(allocator_, &states[1], 2);
auto allocator =
const_cast<OnlineConformerTransducerModel *>(this)->allocator_;
std::vector<Ort::Value> attn_vec = Unbind(allocator, &states[0], 2);
std::vector<Ort::Value> conv_vec = Unbind(allocator, &states[1], 2);
assert(attn_vec.size() == batch_size);
assert(conv_vec.size() == batch_size);