migrate existing set_worker_random_seed callsites

Summary:
**Diff Purpose & Changes**
- We can migrate all instances of worker.set_random_seed_impl to now use builtins.random.set_manual_seed.
- Wonder if some of these cases would benefit from random_seed() instead. Maybe that can be checked out in the future.

Reviewed By: suo

Differential Revision: D72945165

Author: amirafzali

python/monarch/worker/worker.py

@@ -60,10 +60,6 @@
     CONTROLLER_COMPILED_REPEAT = True
 
 
-def set_random_seed_impl(seed, process_idx):
-    torch.manual_seed(seed ^ process_idx)
-
-
 def set_default_dtype(dtype: torch.dtype):
     torch.set_default_dtype(dtype)
 

python/tests/test_remote_functions.py

@@ -30,6 +30,7 @@
 )
 from monarch._testing import BackendType, TestingContext
 from monarch.builtins.log import log_remote
+from monarch.builtins.random import set_manual_seed_remote
 from monarch.cached_remote_function import remote_autograd_function
 from monarch.common import remote as remote_module
 from monarch.common.device_mesh import DeviceMesh
@@ -742,12 +743,8 @@ def something_else():
                 something_else()
 
     def test_setting_random_seed(self, backend_type):
-        set_worker_random_seed = remote(
-            "monarch.worker.worker.set_random_seed_impl", propagate="inspect"
-        )
-
         with self.local_device_mesh(2, 2, backend_type):
-            set_worker_random_seed(12345, 0)
+            set_manual_seed_remote(12345)
             t = torch.randn(3, 4)
             t_d = torch.randn(3, 4, device="cuda")
             ref = fetch_shard(t).result()