[Release] v3.6.1 (#392)

* build(version): v3.6.1

* build(deps): update packages

* style: PLC0415

* style: fix RUF005

* update: beta range to `[0, 1)`

* docs: v3.6.1 changelog

* update: default beta3 range to `[0, 1]`

* update: validate_betas

* update: validate_betas

* update: OPTIMIZERS_IGNORE

* update: test_version_utils

Author: kozistr

docs/changelogs/v3.6.1.md

@@ -1,13 +1,17 @@
 ## Change Log
 
-## Feature
+### Feature
 
 * Implement more cooldown types for WSD learning rate scheduler. (#382, #386)
 * Implement `AdamWSN` optimizer. (#387, #389)
     * [Lean and Mean Adaptive Optimization via Subset-Norm and Subspace-Momentum with Convergence Guarantees](https://arxiv.org/abs/2411.07120)
 * Implement `AdamC` optimizer. (#388, #390)
     * [Why Gradients Rapidly Increase Near the End of Training](https://arxiv.org/abs/2506.02285)
 
+### Update
+
+* Change the default range of the `beta` parameter from `[0, 1]` to `[0, 1)`. (#392)
+
 ### Fix
 
 * Fix to use `momentum buffer` instead of the gradient to calculate LMO. (#385)

examples/visualize_optimizers.py

@@ -16,7 +16,7 @@
 
 filterwarnings('ignore', category=UserWarning)
 
-OPTIMIZERS_IGNORE = ('lomo', 'adalomo', 'demo', 'a2grad', 'muon', 'alice')
+OPTIMIZERS_IGNORE = ('lomo', 'adalomo', 'demo', 'a2grad', 'muon', 'alice', 'adamc', 'adamwsn')
 OPTIMIZERS_MODEL_INPUT_NEEDED = ('lomo', 'adalomo', 'adammini')
 OPTIMIZERS_GRAPH_NEEDED = ('adahessian', 'sophiah')
 OPTIMIZERS_CLOSURE_NEEDED = ('alig', 'bsam')

poetry.lock

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "pytorch_optimizer"
-version = "3.6.0"
+version = "3.6.1"
 description = "optimizer & lr scheduler & objective function collections in PyTorch"
 license = "Apache-2.0"
 authors = ["kozistr <[email protected]>"]
@@ -102,6 +102,7 @@ ignore = [
     "D100", "D102", "D104", "D105", "D107", "D203", "D213", "D413",
     "PLR0912", "PLR0913", "PLR0915", "PLR2004", "PLW2901",
     "Q003", "ARG002",
+    "RUF028",
 ]
 fixable = ["ALL"]
 unfixable = ["F401"]

pyproject.toml

@@ -358,17 +358,17 @@ def validate_mod(x: int, y: int) -> None:
         if x % y != 0:
             raise ValueError(f'[-] {x} must be divisible by {y}')
 
-    def validate_betas(self, betas: BETAS) -> None:
+    def validate_betas(self, betas: BETAS, beta_range_type: str = '[)', beta3_range_type: str = '[]') -> None:
         if betas[0] is not None:
-            self.validate_range(betas[0], 'beta1', 0.0, 1.0, range_type='[]')
+            self.validate_range(betas[0], 'beta1', 0.0, 1.0, range_type=beta_range_type)
 
-        self.validate_range(betas[1], 'beta2', 0.0, 1.0, range_type='[]')
+        self.validate_range(betas[1], 'beta2', 0.0, 1.0, range_type=beta_range_type)
 
         if len(betas) < 3:
             return
 
         if betas[2] is not None:
-            self.validate_range(betas[2], 'beta3', 0.0, 1.0, range_type='[]')
+            self.validate_range(betas[2], 'beta3', 0.0, 1.0, range_type=beta3_range_type)
 
     def validate_nus(self, nus: Union[float, Tuple[float, float]]) -> None:
         if isinstance(nus, float):

pytorch_optimizer/base/optimizer.py

@@ -50,7 +50,7 @@ def compute_normalization_binary_search(activations: torch.Tensor, t: float, num
         activations.dtype
     )
 
-    shape_partition: Tuple[int, ...] = activations.shape[:-1] + (1,)
+    shape_partition: Tuple[int, ...] = (*activations.shape[:-1], 1)
 
     lower = torch.zeros(shape_partition, dtype=activations.dtype, device=activations.device)
     upper = -log_t(1.0 / effective_dim, t) * torch.ones_like(lower)

pytorch_optimizer/loss/bi_tempered.py

@@ -112,7 +112,7 @@
 
 def load_bnb_optimizer(optimizer: str) -> OPTIMIZER:  # pragma: no cover  # noqa: PLR0911
     r"""Load bnb optimizer instance."""
-    from bitsandbytes import optim
+    from bitsandbytes import optim  # noqa: PLC0415
 
     if 'sgd8bit' in optimizer:
         return optim.SGD8bit
@@ -168,7 +168,7 @@ def load_bnb_optimizer(optimizer: str) -> OPTIMIZER:  # pragma: no cover  # noqa
 
 def load_q_galore_optimizer(optimizer: str) -> OPTIMIZER:  # pragma: no cover
     r"""Load Q-GaLore optimizer instance."""
-    import q_galore_torch
+    import q_galore_torch  # noqa: PLC0415
 
     if 'adamw8bit' in optimizer:
         return q_galore_torch.QGaLoreAdamW8bit
@@ -178,7 +178,7 @@ def load_q_galore_optimizer(optimizer: str) -> OPTIMIZER:  # pragma: no cover
 
 def load_ao_optimizer(optimizer: str) -> OPTIMIZER:  # pragma: no cover
     r"""Load TorchAO optimizer instance."""
-    from torchao.prototype import low_bit_optim
+    from torchao.prototype import low_bit_optim  # noqa: PLC0415
 
     if 'adamw8bit' in optimizer:
         return low_bit_optim.AdamW8bit

pytorch_optimizer/optimizer/__init__.py

@@ -33,7 +33,7 @@ def __init__(
         **kwargs,
     ):
         self.validate_learning_rate(lr)
-        self.validate_betas(betas)
+        self.validate_betas(betas, beta_range_type='[]')
         self.validate_non_negative(weight_decay, 'weight_decay')
         self.validate_non_negative(eps, 'eps')
 

pytorch_optimizer/optimizer/pnm.py

@@ -5,7 +5,7 @@ black==25.1.0 ; python_version >= "3.9"
 click==8.1.8 ; python_version >= "3.8"
 colorama==0.4.6 ; python_version >= "3.8" and (sys_platform == "win32" or platform_system == "Windows")
 coverage[toml]==7.6.1 ; python_version == "3.8"
-coverage[toml]==7.8.2 ; python_version >= "3.9"
+coverage[toml]==7.9.2 ; python_version >= "3.9"
 exceptiongroup==1.3.0 ; python_version < "3.11" and python_version >= "3.8"
 filelock==3.16.1 ; python_version == "3.8"
 filelock==3.18.0 ; python_version >= "3.9"
@@ -31,11 +31,12 @@ pluggy==1.5.0 ; python_version == "3.8"
 pluggy==1.6.0 ; python_version >= "3.9"
 pytest-cov==5.0.0 ; python_version >= "3.8"
 pytest==8.3.5 ; python_version >= "3.8"
-ruff==0.11.12 ; python_version >= "3.8"
+ruff==0.12.2 ; python_version >= "3.8"
 setuptools==80.9.0 ; python_version >= "3.12"
 sympy==1.13.3 ; python_version == "3.8"
 sympy==1.14.0 ; python_version >= "3.9"
 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==2.7.0+cpu ; python_version >= "3.9"
-typing-extensions==4.13.2 ; python_version >= "3.8"
+torch==2.7.1+cpu ; python_version >= "3.9"
+typing-extensions==4.13.2 ; python_version == "3.8"
+typing-extensions==4.14.1 ; python_version >= "3.9"

requirements-dev.txt

@@ -16,5 +16,6 @@ setuptools==80.9.0 ; python_version >= "3.12"
 sympy==1.13.3 ; python_version == "3.8"
 sympy==1.14.0 ; python_version >= "3.9"
 torch==2.4.1+cpu ; python_version == "3.8"
-torch==2.7.0+cpu ; python_version >= "3.9"
-typing-extensions==4.13.2 ; python_version >= "3.8"
+torch==2.7.1+cpu ; python_version >= "3.9"
+typing-extensions==4.13.2 ; python_version == "3.8"
+typing-extensions==4.14.1 ; python_version >= "3.9"