Bump torch from 2.3.1+cpu to 2.4.0+cpu

ranzen/hydra/utils.py

@@ -202,10 +202,11 @@ class Config:
                     if (group := groups.get(entry.name)) is not None:
                         for var_name, var_class in group.items():
                             if not issubclass(var_class, typ):  # type: ignore
+                                typ_name = typ.__name__  # type: ignore
                                 raise ValueError(
                                     f"All variants should be subclasses of their entry's type: type"
                                     f" `{var_class.__name__}` of variant `{entry.name}={var_name}` "
-                                    f"is not a subclass of `{typ.__name__}`."
+                                    f"is not a subclass of `{typ_name}`."
                                 )
                     else:
                         raise ValueError(

ranzen/torch/optimizers/adafactor.py

@@ -6,7 +6,7 @@
 
 import torch
 from torch import Tensor
-from torch.optim import Optimizer
+from torch.optim.optimizer import Optimizer
 
 from .common import LossClosure
 

ranzen/torch/schedulers.py

@@ -6,8 +6,8 @@
 
 import torch
 from torch import Tensor
-from torch.optim import Optimizer
 from torch.optim.lr_scheduler import CosineAnnealingLR, _LRScheduler
+from torch.optim.optimizer import Optimizer
 
 __all__ = [
     "CosineLRWithLinearWarmup",
@@ -128,7 +128,7 @@ def scheduler(self) -> Union[LinearWarmupLR, CosineAnnealingLR]:
             self._scheduler = CosineAnnealingLR(
                 optimizer=self.optimizer,
                 T_max=self.total_iters - self.warmup_iters + 1,
-                eta_min=self.lr_min,
+                eta_min=self.lr_min,  # type: ignore
             )
         return self._scheduler
 

tests/optimizers_test.py

@@ -1,7 +1,7 @@
 import pytest
 import torch
 from torch import Tensor
-from torch.optim import AdamW
+from torch.optim.adamw import AdamW
 
 from ranzen.torch.optimizers import Adafactor, LAMB, SAM
 

tests/scheduler_test.py

@@ -1,6 +1,7 @@
 import pytest
 import torch
-from torch import optim
+from torch.optim.adamw import AdamW
+from torch.optim.sgd import SGD
 
 from ranzen.torch.schedulers import LinearWarmupLR
 
@@ -9,7 +10,7 @@ def test_linear_warmup_lr() -> None:
     params = (torch.randn(1, 1, requires_grad=True),)
     base_lr = 1.0
     lr_start = 1.0e-1
-    optimizer = optim.SGD(params, lr=base_lr)
+    optimizer = SGD(params, lr=base_lr)
     scheduler = LinearWarmupLR(optimizer=optimizer, lr_start=lr_start, warmup_iters=1)
     for group in optimizer.param_groups:
         assert group["lr"] == lr_start
@@ -22,7 +23,7 @@ def _step():
     for group in optimizer.param_groups:
         assert group["lr"] == base_lr
 
-    optimizer = optim.AdamW(params, lr=base_lr)
+    optimizer = AdamW(params, lr=base_lr)
     scheduler = LinearWarmupLR(optimizer=optimizer, lr_start=lr_start, warmup_iters=0)
     for group in optimizer.param_groups:
         assert group["lr"] == base_lr
@@ -31,7 +32,7 @@ def _step():
     for group in optimizer.param_groups:
         assert group["lr"] == base_lr
 
-    optimizer = optim.AdamW(params, lr=base_lr)
+    optimizer = AdamW(params, lr=base_lr)
     scheduler = LinearWarmupLR(optimizer=optimizer, lr_start=lr_start, warmup_iters=2)
     _step()
     expected_lr_after_one_step = lr_start + 0.5 * (base_lr - lr_start)