feat: support libtorch

This commit is contained in:
FlamingoPg
2025-10-21 09:55:33 +00:00
parent 239112cb4c
commit 301cbc1d75
22 changed files with 652 additions and 245 deletions
+5
View File
@@ -17,6 +17,11 @@ class DeviceRuntime {
cublasLtHandle_t cublaslt_handle{};
std::shared_ptr<torch::Tensor> cublaslt_workspace;
// cuBLASLt utils
static constexpr size_t kCublasLtWorkspaceSize = 32 * 1024 * 1024;
cublasLtHandle_t cublaslt_handle{};
std::shared_ptr<torch::Tensor> cublaslt_workspace;
public:
explicit DeviceRuntime() {
cublaslt_workspace = std::make_shared<torch::Tensor>(torch::empty({kCublasLtWorkspaceSize}, dtype(torch::kByte).device(at::kCUDA)));