Alphabetize, link to docs for functions

Author: adamchainz

README.rst

@@ -67,9 +67,15 @@ All of these features are on by default but can be disabled with flags.
   .. |numpy.random| replace:: ``numpy.random``
   __ https://numpy.org/doc/stable/reference/random/index.html
 
-* If `TensorFlow <https://www.tensorflow.org/>`_ is installed, its random seed in ``tensorflow.random`` is reset at the start of every test.
+* If `PyTorch <https://pytorch.org/>`_ is installed, its random seed is reset at the start of every test with |torch.manual_seed()|__.
 
-* If `PyTorch <https://pytorch.org/>`_ is installed, its random seed is reset at the start of every test. The random seed of each test is recorded, and can play a role in detecting flaky tests.
+  .. |torch.manual_seed()| replace:: ``torch.manual_seed()``
+  __ https://pytorch.org/docs/stable/generated/torch.manual_seed.html
+
+* If `TensorFlow <https://www.tensorflow.org/>`_ is installed, its random seed is reset at the start of every test with |tensorflow.random.set_seed()|__.
+
+  .. |tensorflow.random.set_seed()| replace:: ``tensorflow.random.set_seed()``
+  __ https://www.tensorflow.org/api_docs/python/tf/random/set_seed
 
 * If additional random generators are used, they can be registered under the
   ``pytest_randomly.random_seeder``

src/pytest_randomly/__init__.py

@@ -67,14 +67,6 @@
 except ImportError:  # pragma: no cover
     have_numpy = False
 
-# tensorflow
-try:
-    import tensorflow as tf
-
-    have_tensorflow = True
-except ImportError:  # pragma: no cover
-    have_tensorflow = False
-
 # pytorch
 try:
     import torch
@@ -83,6 +75,14 @@
 except ImportError:  # pragma: no cover
     have_pytorch = False
 
+# tensorflow
+try:
+    import tensorflow as tf
+
+    have_tensorflow = True
+except ImportError:  # pragma: no cover
+    have_tensorflow = False
+
 
 default_seed = random.Random().getrandbits(32)
 
@@ -196,17 +196,17 @@ def _reseed(config: Config, offset: int = 0) -> int:
         else:
             np_random.set_state(np_random_states[numpy_seed])
 
+    if have_pytorch:  # pragma: no branch
+        torch.manual_seed(seed)
+        if torch.cuda.is_available():  # Also seed CUDA if available
+            torch.cuda.manual_seed_all(seed)
+
     if have_tensorflow:  # pragma: no branch
         tf.random.set_seed(seed)
         # TensorFlow 1.x compatibility
         if hasattr(tf, "compat"):
             tf.compat.v1.set_random_seed(seed)
 
-    if have_pytorch:  # pragma: no branch
-        torch.manual_seed(seed)
-        if torch.cuda.is_available():  # Also seed CUDA if available
-            torch.cuda.manual_seed_all(seed)
-
     if entrypoint_reseeds is None:
         eps = entry_points(group="pytest_randomly.random_seeder")
         entrypoint_reseeds = [e.load() for e in eps]