diff --git a/pytorch3d/common/compat.py b/pytorch3d/common/compat.py
index f2edde0e..5f25a185 100644
--- a/pytorch3d/common/compat.py
+++ b/pytorch3d/common/compat.py
@@ -19,7 +19,7 @@ def solve(A: torch.Tensor, B: torch.Tensor) -> torch.Tensor:  # pragma: no cover
     Like torch.linalg.solve, tries to return X
     such that AX=B, with A square.
     """
-    if hasattr(torch.linalg, "solve"):
+    if hasattr(torch, "linalg") and hasattr(torch.linalg, "solve"):
         # PyTorch version >= 1.8.0
         return torch.linalg.solve(A, B)
 
@@ -31,7 +31,7 @@ def lstsq(A: torch.Tensor, B: torch.Tensor) -> torch.Tensor:  # pragma: no cover
     Like torch.linalg.lstsq, tries to return X
     such that AX=B.
     """
-    if hasattr(torch.linalg, "lstsq"):
+    if hasattr(torch, "linalg") and hasattr(torch.linalg, "lstsq"):
         # PyTorch version >= 1.9
         return torch.linalg.lstsq(A, B).solution
 
@@ -45,7 +45,7 @@ def qr(A: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]:  # pragma: no cove
     """
     Like torch.linalg.qr.
     """
-    if hasattr(torch.linalg, "qr"):
+    if hasattr(torch, "linalg") and hasattr(torch.linalg, "qr"):
         # PyTorch version >= 1.9
         return torch.linalg.qr(A)
     return torch.qr(A)
diff --git a/pytorch3d/renderer/cameras.py b/pytorch3d/renderer/cameras.py
index b36e1cee..217974d6 100644
--- a/pytorch3d/renderer/cameras.py
+++ b/pytorch3d/renderer/cameras.py
@@ -6,7 +6,7 @@
 
 import math
 import warnings
-from typing import Optional, Sequence, Tuple, Union, List
+from typing import List, Optional, Sequence, Tuple, Union
 
 import numpy as np
 import torch
@@ -259,8 +259,9 @@ class CamerasBase(TensorProperties):
             # users might might have to implement the screen to NDC transform based
             # on the definition of the camera parameters.
             # See PerspectiveCameras/OrthographicCameras for an example.
-            # We don't flip xy because we assume that world points are in PyTorch3D coodrinates
-            # and thus conversion from screen to ndc is a mere scaling from image to [-1, 1] scale.
+            # We don't flip xy because we assume that world points are in
+            # PyTorch3D coordinates, and thus conversion from screen to ndc
+            # is a mere scaling from image to [-1, 1] scale.
             return get_screen_to_ndc_transform(self, with_xyflip=False, **kwargs)
 
     def transform_points_ndc(
diff --git a/pytorch3d/renderer/points/pulsar/unified.py b/pytorch3d/renderer/points/pulsar/unified.py
index 8ed5980b..5c1fb51d 100644
--- a/pytorch3d/renderer/points/pulsar/unified.py
+++ b/pytorch3d/renderer/points/pulsar/unified.py
@@ -551,17 +551,15 @@ class PulsarPointsRenderer(nn.Module):
                 otherargs["bg_col"] = bg_col
             # Go!
             images.append(
-                torch.flipud(
-                    self.renderer(
-                        vert_pos=vert_pos,
-                        vert_col=vert_col,
-                        vert_rad=vert_rad,
-                        cam_params=cam_params,
-                        gamma=gamma,
-                        max_depth=zfar,
-                        min_depth=znear,
-                        **otherargs,
-                    )
-                )
+                self.renderer(
+                    vert_pos=vert_pos,
+                    vert_col=vert_col,
+                    vert_rad=vert_rad,
+                    cam_params=cam_params,
+                    gamma=gamma,
+                    max_depth=zfar,
+                    min_depth=znear,
+                    **otherargs,
+                ).flip(dims=[0])
             )
         return torch.stack(images, dim=0)
diff --git a/pytorch3d/transforms/rotation_conversions.py b/pytorch3d/transforms/rotation_conversions.py
index 396eaa18..c3d01488 100644
--- a/pytorch3d/transforms/rotation_conversions.py
+++ b/pytorch3d/transforms/rotation_conversions.py
@@ -140,8 +140,10 @@ def matrix_to_quaternion(matrix: torch.Tensor) -> torch.Tensor:
         dim=-2,
     )
 
-    # clipping is not important here; if q_abs is small, the candidate won't be picked
-    quat_candidates = quat_by_rijk / (2.0 * q_abs[..., None].clip(0.1))
+    # We floor here at 0.1 but the exact level is not important; if q_abs is small,
+    # the candidate won't be picked.
+    # pyre-ignore [16]: `torch.Tensor` has no attribute `new_tensor`.
+    quat_candidates = quat_by_rijk / (2.0 * q_abs[..., None].max(q_abs.new_tensor(0.1)))
 
     # if not for numerical problems, quat_candidates[i] should be same (up to a sign),
     # forall i; we pick the best-conditioned one (with the largest denominator)