Little cleanup in so3.py and _so3.py (#231)

* model version for Potential class is added

* model version for Potential class is modified

* Enable the smooth version of Spherical Bessel function in TensorNet

* max_n, max_l for SphericalBessel radial basis functions are included in TensorNet class

* adding united tests for improving the coverage score

* little clean up in _so3.py and so3.py

src/matgl/layers/_so3.py

@@ -66,8 +66,8 @@ def __init__(self, lmax: int):
 
         ls = torch.arange(0, lmax + 1)
         nls = 2 * ls + 1
-        self.lidx = torch.repeat_interleave(ls, nls).cpu()
-        self.midx = torch.cat([torch.arange(-l_id, l_id + 1) for l_id in ls]).cpu()
+        self.lidx = torch.repeat_interleave(ls, nls)
+        self.midx = torch.cat([torch.arange(-l_id, l_id + 1) for l_id in ls])
 
         self.register_buffer("flidx", self.lidx.to(dtype=dtype), False)
 
@@ -356,7 +356,6 @@ def __init__(self, n_in: int, lmax: int):
         self.n_in = n_in
         self.lidx, _ = sh_indices(lmax)
         self.scaling = nn.Linear(n_in, n_in * (lmax + 1))
-        self.lidx = self.lidx.cpu()  # added by kenko
 
     def forward(self, x: torch.Tensor) -> torch.Tensor:
         s0 = x[:, 0, :]

src/matgl/utils/so3.py

@@ -61,8 +61,8 @@ def generate_clebsch_gordan(lmax: int) -> torch.Tensor:
     """
     lidx, midx = sh_indices(lmax)
     cg = torch.zeros((lidx.shape[0], lidx.shape[0], lidx.shape[0]))
-    lidx = lidx.cpu().numpy()
-    midx = midx.cpu().numpy()
+    lidx = lidx.numpy()
+    midx = midx.numpy()
     for c1, (l1, m1) in enumerate(zip(lidx, midx)):
         for c2, (l2, m2) in enumerate(zip(lidx, midx)):
             for c3, (l3, m3) in enumerate(zip(lidx, midx)):