Remove unnecessary cholesky_jitter call (#929)

Summary:
I don't think this is doing anything; it seems to be a nullop?

Pull Request resolved: #929

Reviewed By: wjmaddox

Differential Revision: D30729615

Pulled By: Balandat

fbshipit-source-id: 0df4694c97a526b9b09a0a67807b1dd6c19f56f2

Author: Balandat

botorch/models/multitask.py

@@ -63,7 +63,7 @@
 from gpytorch.priors.prior import Prior
 from gpytorch.priors.smoothed_box_prior import SmoothedBoxPrior
 from gpytorch.priors.torch_priors import GammaPrior
-from gpytorch.settings import cholesky_jitter, detach_test_caches
+from gpytorch.settings import detach_test_caches
 from gpytorch.utils.errors import CachingError
 from gpytorch.utils.memoize import cached, pop_from_cache
 from torch import Tensor
@@ -571,15 +571,14 @@ def posterior(
 
         # now update root so that \tilde{R}\tilde{R}' \approx K_{(x,xt), (x,xt)}
         # cloning preserves the gradient history
-        with cholesky_jitter(cholesky_jitter.value(X.dtype)):
-            updated_lazy_tensor = data_data_covar.cat_rows(
-                cross_mat=test_data_covar.clone(),
-                new_mat=test_test_covar,
-                method="diagonalization",
-            )
-            updated_root = updated_lazy_tensor.root_decomposition().root
-            # occasionally, there's device errors so enforce this comes out right
-            updated_root = updated_root.to(data_data_covar.device)
+        updated_lazy_tensor = data_data_covar.cat_rows(
+            cross_mat=test_data_covar.clone(),
+            new_mat=test_test_covar,
+            method="diagonalization",
+        )
+        updated_root = updated_lazy_tensor.root_decomposition().root
+        # occasionally, there's device errors so enforce this comes out right
+        updated_root = updated_root.to(data_data_covar.device)
 
         # build a root decomposition of the joint train/test covariance matrix
         # construct (\tilde{R} \otimes M)(\tilde{R} \otimes M)' \approx