Merge pull request #374 from kozistr/release/v3.5.1

[Release] v3.5.1

Author: kozistr

poetry.lock

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "pytorch_optimizer"
-version = "3.5.0"
+version = "3.5.1"
 description = "optimizer & lr scheduler & objective function collections in PyTorch"
 license = "Apache-2.0"
 authors = ["kozistr <[email protected]>"]

pyproject.toml

@@ -31,7 +31,8 @@ def step(self, current_step: int) -> None:
 
         :param current_step: int. Current step index.
         """
-        self.cosine_stepper.step(current_step)
+        self.cosine_stepper.last_epoch = current_step
+        self.cosine_stepper.step()
 
     def get_death_rate(self, current_step: int) -> float:
         r"""Get the updated rate (death_rate) at the given step.
@@ -266,9 +267,9 @@ class StableSPAM(BaseOptimizer):
     :param params: PARAMETERS. iterable of parameters to optimize or dicts defining parameter groups.
     :param lr: float. learning rate.
     :param betas: BETAS. coefficients used for computing running averages of gradient and the squared hessian trace.
-    :param gamma1: float.
-    :param gamma2: float.
-    :param theta: float.
+    :param gamma1: float. gamma1 parameter.
+    :param gamma2: float. gamma2 parameter.
+    :param theta: float. theta parameter.
     :param t_max: Optional[int]. total number of steps.
     :param eta_min: float. eta_min of CosineDecay.
     :param weight_decay: float. weight decay (L2 penalty).

pytorch_optimizer/optimizer/spam.py

@@ -18,7 +18,7 @@ jinja2==3.1.6 ; python_version >= "3.8"
 markupsafe==2.1.5 ; python_version == "3.8"
 markupsafe==3.0.2 ; python_version >= "3.9"
 mpmath==1.3.0 ; python_version >= "3.8"
-mypy-extensions==1.0.0 ; python_version >= "3.8"
+mypy-extensions==1.1.0 ; python_version >= "3.8"
 networkx==3.1 ; python_version == "3.8"
 networkx==3.2.1 ; python_version >= "3.9"
 numpy==1.24.4 ; python_version == "3.8"
@@ -30,11 +30,10 @@ platformdirs==4.3.7 ; python_version >= "3.9"
 pluggy==1.5.0 ; python_version >= "3.8"
 pytest-cov==5.0.0 ; python_version >= "3.8"
 pytest==8.3.5 ; python_version >= "3.8"
-ruff==0.11.6 ; python_version >= "3.8"
-setuptools==79.0.0 ; python_version >= "3.12"
-sympy==1.13.1 ; python_version >= "3.9"
-sympy==1.13.3 ; python_version == "3.8"
+ruff==0.11.7 ; python_version >= "3.8"
+setuptools==79.0.1 ; python_version >= "3.12"
+sympy==1.13.3 ; python_version >= "3.8"
 tomli==2.2.1 ; python_full_version <= "3.11.0a6" and python_version >= "3.8"
 torch==2.4.1+cpu ; python_version == "3.8"
-torch==2.6.0+cpu ; python_version >= "3.9"
+torch==2.7.0+cpu ; python_version >= "3.9"
 typing-extensions==4.13.2 ; python_version >= "3.8"

requirements-dev.txt

@@ -12,9 +12,8 @@ networkx==3.1 ; python_version == "3.8"
 networkx==3.2.1 ; python_version >= "3.9"
 numpy==1.24.4 ; python_version == "3.8"
 numpy==2.0.2 ; python_version >= "3.9"
-setuptools==79.0.0 ; python_version >= "3.12"
-sympy==1.13.1 ; python_version >= "3.9"
-sympy==1.13.3 ; python_version == "3.8"
+setuptools==79.0.1 ; python_version >= "3.12"
+sympy==1.13.3 ; python_version >= "3.8"
 torch==2.4.1+cpu ; python_version == "3.8"
-torch==2.6.0+cpu ; python_version >= "3.9"
+torch==2.7.0+cpu ; python_version >= "3.9"
 typing-extensions==4.13.2 ; python_version >= "3.8"

requirements.txt

@@ -200,17 +200,19 @@ def test_get_chebyshev_lr():
     optimizer.step()
 
     lr_scheduler = get_chebyshev_schedule(optimizer, num_epochs=16, is_warmup=True)
-    lr_scheduler.step(0)
+    lr_scheduler.last_epoch = 0
+    lr_scheduler.step()
 
     np.testing.assert_almost_equal(lr_scheduler.get_last_lr(), 1e-3)
 
     optimizer = AdamW(Example().parameters())
     optimizer.step()
 
     lr_scheduler = get_chebyshev_schedule(optimizer, num_epochs=16, is_warmup=False)
+    lr_scheduler.last_epoch = 0
 
-    for i, expected_lr in enumerate(recipes, start=1):
-        lr_scheduler.step(i)
+    for expected_lr in recipes:
+        lr_scheduler.step()
         np.testing.assert_almost_equal(lr_scheduler.get_last_lr(), expected_lr)
 
 
@@ -311,10 +313,10 @@ def test_wsd_lr_scheduler():
 
     lr_scheduler = get_wsd_schedule(optimizer, 2, 2, 3, min_lr_ratio=0.1)
 
-    expected_lrs = [0.0, 0.0005, 0.001, 0.001, 0.001, 0.000775, 0.000325, 0.0001, 0.0001, 0.0001]
+    expected_lrs = [0.0005, 0.001, 0.001, 0.001, 0.000775, 0.000325, 0.0001, 0.0001, 0.0001]
 
-    for step, expected_lr in enumerate(expected_lrs):
-        lr_scheduler.step(step)
+    for expected_lr in expected_lrs:
+        lr_scheduler.step()
         np.testing.assert_almost_equal(expected_lr, lr_scheduler.get_last_lr(), 6)
 
 

tests/test_lr_schedulers.py

@@ -247,9 +247,9 @@ def test_version_utils():
     with pytest.raises(ValueError):
         parse_pytorch_version('a.s.d.f')
 
-    assert parse_pytorch_version(torch.__version__) == [2, 6, 0]
+    assert parse_pytorch_version(torch.__version__) == [2, 7, 0]
 
-    assert compare_versions('2.6.0', '2.4.0') >= 0
+    assert compare_versions('2.7.0', '2.4.0') >= 0
 
 
 def test_cpu_offload_optimizer():