some better imports and fixes

crates/burn-core/src/backend.rs

@@ -29,4 +29,4 @@ pub use burn_tch as libtorch;
 pub use burn_tch::LibTorch;
 
 #[cfg(feature = "sparse")]
-pub use burn_sparse as sparse;
+pub use burn_sparse::decorator as sparse;

crates/burn-core/src/tensor.rs

@@ -1 +1,6 @@
 pub use burn_tensor::*;
+
+#[cfg(feature = "sparse")]
+pub mod sparse {
+    pub use burn_sparse::backend::*;
+}

crates/burn-sparse/src/backend/api.rs

@@ -1,7 +1,14 @@
 use crate::backend::{Sparse, SparseBackend};
 use burn_tensor::{Int, Tensor, TensorPrimitive};
 
-pub trait SparseTensor<const D: usize, B>
+pub trait ToSparse<const D: usize, B>
+where
+    B: SparseBackend,
+{
+    fn into_sparse(self) -> Tensor<B, D, Sparse>;
+}
+
+pub trait SparseTensorApi<const D: usize, B>
 where
     B: SparseBackend,
 {
@@ -10,7 +17,16 @@ where
     fn dense(self) -> Tensor<B, D>;
 }
 
-impl<const D: usize, B> SparseTensor<D, B> for Tensor<B, D, Sparse>
+impl<const D: usize, B> ToSparse<D, B> for Tensor<B, D>
+where
+    B: SparseBackend,
+{
+    fn into_sparse(self) -> Tensor<B, D, Sparse> {
+        Tensor::new(B::sparse_to_sparse(self.into_primitive().tensor()))
+    }
+}
+
+impl<const D: usize, B> SparseTensorApi<D, B> for Tensor<B, D, Sparse>
 where
     B: SparseBackend,
 {

crates/burn-sparse/src/backend/mod.rs

@@ -4,5 +4,6 @@ mod kind;
 mod sparse_backend;
 
 pub use alias::*;
+pub use api::*;
 pub use kind::*;
 pub use sparse_backend::*;