Add files using upload-large-folder tool

pytorch-image-models/timm/optim/__init__.py

@@ -0,0 +1,34 @@
+from .adabelief import AdaBelief
+from .adafactor import Adafactor
+from .adafactor_bv import AdafactorBigVision
+from .adahessian import Adahessian
+from .adamp import AdamP
+from .adamw import AdamWLegacy
+from .adan import Adan
+from .adopt import Adopt
+from .lamb import Lamb
+from .laprop import LaProp
+from .lars import Lars
+from .lion import Lion
+from .lookahead import Lookahead
+from .madgrad import MADGRAD
+from .mars import Mars
+from .nadam import NAdamLegacy
+from .nadamw import NAdamW
+from .nvnovograd import NvNovoGrad
+from .radam import RAdamLegacy
+from .rmsprop_tf import RMSpropTF
+from .sgdp import SGDP
+from .sgdw import SGDW
+
+# bring common torch.optim Optimizers into timm.optim namespace for consistency
+from torch.optim import Adadelta, Adagrad, Adamax, Adam, AdamW, RMSprop, SGD
+try:
+    # in case any very old torch versions being used
+    from torch.optim import NAdam, RAdam
+except ImportError:
+    pass
+
+from ._optim_factory import list_optimizers, get_optimizer_class, get_optimizer_info, OptimInfo, OptimizerRegistry, \
+    create_optimizer_v2, create_optimizer, optimizer_kwargs
+from ._param_groups import param_groups_layer_decay, param_groups_weight_decay, auto_group_layers

pytorch-image-models/timm/optim/_optim_factory.py

@@ -0,0 +1,1081 @@
+""" Optimizer Factory w/ custom Weight Decay & Layer Decay support
+
+Hacked together by / Copyright 2021 Ross Wightman
+"""
+import logging
+from dataclasses import dataclass
+from functools import partial
+from typing import Any, Callable, Dict, List, Optional, Set, Tuple, Type, Union
+from fnmatch import fnmatch
+import importlib
+
+import torch
+import torch.nn as nn
+import torch.optim
+
+from ._param_groups import param_groups_layer_decay, param_groups_weight_decay
+from ._types import ParamsT, OptimType, OptimizerCallable
+from .adabelief import AdaBelief
+from .adafactor import Adafactor
+from .adafactor_bv import AdafactorBigVision
+from .adahessian import Adahessian
+from .adamp import AdamP
+from .adamw import AdamWLegacy
+from .adan import Adan
+from .adopt import Adopt
+from .lamb import Lamb
+from .laprop import LaProp
+from .lars import Lars
+from .lion import Lion
+from .lookahead import Lookahead
+from .madgrad import MADGRAD
+from .mars import Mars
+from .nadam import NAdamLegacy
+from .nadamw import NAdamW
+from .nvnovograd import NvNovoGrad
+from .radam import RAdamLegacy
+from .rmsprop_tf import RMSpropTF
+from .sgdp import SGDP
+from .sgdw import SGDW
+
+_logger = logging.getLogger(__name__)
+
+
+def _import_class(class_string: str) -> Type:
+    """Dynamically import a class from a string."""
+    try:
+        module_name, class_name = class_string.rsplit(".", 1)
+        module = importlib.import_module(module_name)
+        return getattr(module, class_name)
+    except (ImportError, AttributeError) as e:
+        raise ImportError(f"Could not import {class_string}: {e}")
+
+
+
+@dataclass(frozen=True)
+class OptimInfo:
+    """Immutable configuration for an optimizer.
+
+    Attributes:
+        name: Unique identifier for the optimizer
+        opt_class: The optimizer class
+        description: Brief description of the optimizer's characteristics and behavior
+        has_eps: Whether the optimizer accepts epsilon parameter
+        has_momentum: Whether the optimizer accepts momentum parameter
+        has_betas: Whether the optimizer accepts a tuple of beta parameters
+        num_betas: number of betas in tuple (valid IFF has_betas = True)
+        defaults: Optional default parameters for the optimizer
+    """
+    name: str
+    opt_class: Union[str, OptimType]
+    description: str = ''
+    has_eps: bool = True
+    has_momentum: bool = False
+    has_betas: bool = False
+    num_betas: int = 2
+    second_order: bool = False
+    defaults: Optional[Dict[str, Any]] = None
+
+
+class OptimizerRegistry:
+    """Registry managing optimizer configurations and instantiation.
+
+    This class provides a central registry for optimizer configurations and handles
+    their instantiation with appropriate parameter groups and settings.
+    """
+
+    def __init__(self) -> None:
+        self._optimizers: Dict[str, OptimInfo] = {}
+        self._foreach_defaults: Set[str] = {'lion'}
+
+    def register(self, info: OptimInfo) -> None:
+        """Register an optimizer configuration.
+
+        Args:
+            info: The OptimInfo configuration containing name, type and description
+        """
+        name = info.name.lower()
+        if name in self._optimizers:
+            _logger.warning(f'Optimizer {name} already registered, overwriting')
+        self._optimizers[name] = info
+
+    def register_alias(self, alias: str, target: str) -> None:
+        """Register an alias for an existing optimizer.
+
+        Args:
+            alias: The alias name
+            target: The target optimizer name
+
+        Raises:
+            KeyError: If target optimizer doesn't exist
+        """
+        target = target.lower()
+        if target not in self._optimizers:
+            raise KeyError(f'Cannot create alias for non-existent optimizer {target}')
+        self._optimizers[alias.lower()] = self._optimizers[target]
+
+    def register_foreach_default(self, name: str) -> None:
+        """Register an optimizer as defaulting to foreach=True."""
+        self._foreach_defaults.add(name.lower())
+
+    def list_optimizers(
+            self,
+            filter: Union[str, List[str]] = '',
+            exclude_filters: Optional[List[str]] = None,
+            with_description: bool = False
+    ) -> List[Union[str, Tuple[str, str]]]:
+        """List available optimizer names, optionally filtered.
+
+        Args:
+            filter: Wildcard style filter string (e.g., 'adam*')
+            exclude_filters: Optional list of wildcard patterns to exclude
+            with_description: If True, return tuples of (name, description)
+
+        Returns:
+            List of either optimizer names or (name, description) tuples
+        """
+        names = sorted(self._optimizers.keys())
+
+        if filter:
+            if isinstance(filter, str):
+                filters = [filter]
+            else:
+                filters = filter
+            filtered_names = set()
+            for f in filters:
+                filtered_names.update(n for n in names if fnmatch(n, f))
+            names = sorted(filtered_names)
+
+        if exclude_filters:
+            for exclude_filter in exclude_filters:
+                names = [n for n in names if not fnmatch(n, exclude_filter)]
+
+        if with_description:
+            return [(name, self._optimizers[name].description) for name in names]
+
+        return names
+
+    def get_optimizer_info(self, name: str) -> OptimInfo:
+        """Get the OptimInfo for an optimizer.
+
+        Args:
+            name: Name of the optimizer
+
+        Returns:
+            OptimInfo configuration
+
+        Raises:
+            ValueError: If optimizer is not found
+        """
+        name = name.lower()
+        if name not in self._optimizers:
+            raise ValueError(f'Optimizer {name} not found in registry')
+        return self._optimizers[name]
+
+    def get_optimizer_class(
+            self,
+            name_or_info: Union[str, OptimInfo],
+            bind_defaults: bool = True,
+    ) -> Union[OptimType, OptimizerCallable]:
+        """Get the optimizer class with any default arguments applied.
+
+        This allows direct instantiation of optimizers with their default configs
+        without going through the full factory.
+
+        Args:
+            name_or_info: Name of the optimizer
+            bind_defaults: Bind default arguments to optimizer class via `partial` before returning
+
+        Returns:
+            Optimizer class or partial with defaults applied
+
+        Raises:
+            ValueError: If optimizer not found
+        """
+        if isinstance(name_or_info, str):
+            opt_info = self.get_optimizer_info(name_or_info)
+        else:
+            assert isinstance(name_or_info, OptimInfo)
+            opt_info = name_or_info
+
+        if isinstance(opt_info.opt_class, str):
+            # Special handling for APEX and BNB optimizers
+            if opt_info.opt_class.startswith('apex.'):
+                assert torch.cuda.is_available(), 'CUDA required for APEX optimizers'
+                try:
+                    opt_class = _import_class(opt_info.opt_class)
+                except ImportError as e:
+                    raise ImportError('APEX optimizers require apex to be installed') from e
+            elif opt_info.opt_class.startswith('bitsandbytes.'):
+                assert torch.cuda.is_available(), 'CUDA required for bitsandbytes optimizers'
+                try:
+                    opt_class = _import_class(opt_info.opt_class)
+                except ImportError as e:
+                    raise ImportError('bitsandbytes optimizers require bitsandbytes to be installed') from e
+            else:
+                opt_class = _import_class(opt_info.opt_class)
+        else:
+            opt_class = opt_info.opt_class
+
+        # Return class or partial with defaults
+        if bind_defaults and opt_info.defaults:
+            opt_class = partial(opt_class, **opt_info.defaults)
+
+        return opt_class
+
+    def create_optimizer(
+            self,
+            model_or_params: Union[nn.Module, ParamsT],
+            opt: str,
+            lr: Optional[float] = None,
+            weight_decay: float = 0.,
+            momentum: float = 0.9,
+            foreach: Optional[bool] = None,
+            weight_decay_exclude_1d: bool = True,
+            layer_decay: Optional[float] = None,
+            param_group_fn: Optional[Callable[[nn.Module], ParamsT]] = None,
+            **kwargs: Any,
+    ) -> torch.optim.Optimizer:
+        """Create an optimizer instance.
+
+        Args:
+            model_or_params: Model or parameters to optimize
+            opt: Name of optimizer to create
+            lr: Learning rate
+            weight_decay: Weight decay factor
+            momentum: Momentum factor for applicable optimizers
+            foreach: Enable/disable foreach operation
+            weight_decay_exclude_1d: Whether to skip weight decay for 1d params (biases and norm affine)
+            layer_decay: Layer-wise learning rate decay
+            param_group_fn: Optional custom parameter grouping function
+            **kwargs: Additional optimizer-specific arguments
+
+        Returns:
+            Configured optimizer instance
+
+        Raises:
+            ValueError: If optimizer not found or configuration invalid
+        """
+
+        # Get parameters to optimize
+        if isinstance(model_or_params, nn.Module):
+            # Extract parameters from a nn.Module, build param groups w/ weight-decay and/or layer-decay applied
+            no_weight_decay = getattr(model_or_params, 'no_weight_decay', lambda: set())()
+
+            if param_group_fn:
+                # run custom fn to generate param groups from nn.Module
+                params = param_group_fn(model_or_params)
+            elif layer_decay is not None:
+                params = param_groups_layer_decay(
+                    model_or_params,
+                    weight_decay=weight_decay,
+                    layer_decay=layer_decay,
+                    no_weight_decay_list=no_weight_decay,
+                    weight_decay_exclude_1d=weight_decay_exclude_1d,
+                )
+                weight_decay = 0.
+            elif weight_decay and weight_decay_exclude_1d:
+                params = param_groups_weight_decay(
+                    model_or_params,
+                    weight_decay=weight_decay,
+                    no_weight_decay_list=no_weight_decay,
+                )
+                weight_decay = 0.
+            else:
+                params = model_or_params.parameters()
+        else:
+            # pass parameters / parameter groups through to optimizer
+            params = model_or_params
+
+        # Parse optimizer name
+        opt_split = opt.lower().split('_')
+        opt_name = opt_split[-1]
+        use_lookahead = opt_split[0] == 'lookahead' if len(opt_split) > 1 else False
+
+        opt_info = self.get_optimizer_info(opt_name)
+
+        # Build optimizer arguments
+        opt_args: Dict[str, Any] = {'weight_decay': weight_decay, **kwargs}
+
+        # Add LR to args, if None optimizer default is used, some optimizers manage LR internally if None.
+        if lr is not None:
+            opt_args['lr'] = lr
+
+        # Apply optimizer-specific settings
+        if opt_info.defaults:
+            for k, v in opt_info.defaults.items():
+                opt_args.setdefault(k, v)
+
+        # timm has always defaulted momentum to 0.9 if optimizer supports momentum, keep for backward compat.
+        if opt_info.has_momentum:
+            opt_args.setdefault('momentum', momentum)
+
+        # Remove commonly used kwargs that aren't always supported
+        if not opt_info.has_eps:
+            opt_args.pop('eps', None)
+        if not opt_info.has_betas:
+            opt_args.pop('betas', None)
+
+        if foreach is not None:
+            # Explicitly activate or deactivate multi-tensor foreach impl.
+            # Not all optimizers support this, and those that do usually default to using
+            # multi-tensor impl if foreach is left as default 'None' and can be enabled.
+            opt_args.setdefault('foreach', foreach)
+
+        # Create optimizer
+        opt_class = self.get_optimizer_class(opt_info, bind_defaults=False)
+        optimizer = opt_class(params, **opt_args)
+
+        # Apply Lookahead if requested
+        if use_lookahead:
+            optimizer = Lookahead(optimizer)
+
+        return optimizer
+
+
+def _register_sgd_variants(registry: OptimizerRegistry) -> None:
+    """Register SGD-based optimizers"""
+    sgd_optimizers = [
+        OptimInfo(
+            name='sgd',
+            opt_class=torch.optim.SGD,
+            description='torch.Optim Stochastic Gradient Descent (SGD) with Nesterov momentum',
+            has_eps=False,
+            has_momentum=True,
+            defaults={'nesterov': True}
+        ),
+        OptimInfo(
+            name='momentum',
+            opt_class=torch.optim.SGD,
+            description='torch.Optim Stochastic Gradient Descent (SGD) with classical momentum',
+            has_eps=False,
+            has_momentum=True,
+            defaults={'nesterov': False}
+        ),
+        OptimInfo(
+            name='sgdp',
+            opt_class=SGDP,
+            description='SGD with built-in projection to unit norm sphere',
+            has_momentum=True,
+            defaults={'nesterov': True}
+        ),
+        OptimInfo(
+            name='sgdw',
+            opt_class=SGDW,
+            description='SGD with decoupled weight decay and Nesterov momentum',
+            has_eps=False,
+            has_momentum=True,
+            defaults={'nesterov': True}
+        ),
+    ]
+    for opt in sgd_optimizers:
+        registry.register(opt)
+
+
+def _register_adam_variants(registry: OptimizerRegistry) -> None:
+    """Register Adam-based optimizers"""
+    adam_optimizers = [
+        OptimInfo(
+            name='adam',
+            opt_class=torch.optim.Adam,
+            description='torch.optim.Adam, Adaptive Moment Estimation',
+            has_betas=True
+        ),
+        OptimInfo(
+            name='adamw',
+            opt_class=torch.optim.AdamW,
+            description='torch.optim.AdamW, Adam with decoupled weight decay',
+            has_betas=True
+        ),
+        OptimInfo(
+            name='adamwlegacy',
+            opt_class=AdamWLegacy,
+            description='legacy impl of AdamW that pre-dates inclusion to torch.optim',
+            has_betas=True
+        ),
+        OptimInfo(
+            name='adamp',
+            opt_class=AdamP,
+            description='Adam with built-in projection to unit norm sphere',
+            has_betas=True,
+            defaults={'wd_ratio': 0.01, 'nesterov': True}
+        ),
+        OptimInfo(
+            name='nadam',
+            opt_class=torch.optim.NAdam,
+            description='torch.optim.NAdam, Adam with Nesterov momentum',
+            has_betas=True
+        ),
+        OptimInfo(
+            name='nadamlegacy',
+            opt_class=NAdamLegacy,
+            description='legacy impl of NAdam that pre-dates inclusion in torch.optim',
+            has_betas=True
+        ),
+        OptimInfo(
+            name='nadamw',
+            opt_class=NAdamW,
+            description='Adam with Nesterov momentum and decoupled weight decay, mlcommons/algorithmic-efficiency impl',
+            has_betas=True
+        ),
+        OptimInfo(
+            name='radam',
+            opt_class=torch.optim.RAdam,
+            description='torch.optim.RAdam, Rectified Adam with variance adaptation',
+            has_betas=True
+        ),
+        OptimInfo(
+            name='radamlegacy',
+            opt_class=RAdamLegacy,
+            description='legacy impl of RAdam that predates inclusion in torch.optim',
+            has_betas=True
+        ),
+        OptimInfo(
+            name='radamw',
+            opt_class=torch.optim.RAdam,
+            description='torch.optim.RAdamW, Rectified Adam with variance adaptation and decoupled weight decay',
+            has_betas=True,
+            defaults={'decoupled_weight_decay': True}
+        ),
+        OptimInfo(
+            name='adamax',
+            opt_class=torch.optim.Adamax,
+            description='torch.optim.Adamax, Adam with infinity norm for more stable updates',
+            has_betas=True
+        ),
+        OptimInfo(
+            name='adafactor',
+            opt_class=Adafactor,
+            description='Memory-efficient implementation of Adam with factored gradients',
+        ),
+        OptimInfo(
+            name='adafactorbv',
+            opt_class=AdafactorBigVision,
+            description='Big Vision variant of Adafactor with factored gradients, half precision momentum',
+        ),
+        OptimInfo(
+            name='adopt',
+            opt_class=Adopt,
+            description='Modified Adam that can converge with any β2 with the optimal rate',
+        ),
+        OptimInfo(
+            name='adoptw',
+            opt_class=Adopt,
+            description='Modified AdamW (decoupled decay) that can converge with any β2 with the optimal rate',
+            defaults={'decoupled': True}
+        ),
+    ]
+    for opt in adam_optimizers:
+        registry.register(opt)
+
+
+def _register_lamb_lars(registry: OptimizerRegistry) -> None:
+    """Register LAMB and LARS variants"""
+    lamb_lars_optimizers = [
+        OptimInfo(
+            name='lamb',
+            opt_class=Lamb,
+            description='Layer-wise Adaptive Moments for batch optimization',
+            has_betas=True
+        ),
+        OptimInfo(
+            name='lambc',
+            opt_class=Lamb,
+            description='LAMB with trust ratio clipping for stability',
+            has_betas=True,
+            defaults={'trust_clip': True}
+        ),
+        OptimInfo(
+            name='lars',
+            opt_class=Lars,
+            description='Layer-wise Adaptive Rate Scaling',
+            has_momentum=True
+        ),
+        OptimInfo(
+            name='larc',
+            opt_class=Lars,
+            description='LARS with trust ratio clipping for stability',
+            has_momentum=True,
+            defaults={'trust_clip': True}
+        ),
+        OptimInfo(
+            name='nlars',
+            opt_class=Lars,
+            description='LARS with Nesterov momentum',
+            has_momentum=True,
+            defaults={'nesterov': True}
+        ),
+        OptimInfo(
+            name='nlarc',
+            opt_class=Lars,
+            description='LARS with Nesterov momentum & trust ratio clipping',
+            has_momentum=True,
+            defaults={'nesterov': True, 'trust_clip': True}
+        ),
+    ]
+    for opt in lamb_lars_optimizers:
+        registry.register(opt)
+
+
+def _register_cautious_optimizers(registry: OptimizerRegistry) -> None:
+    cautious_optimizers = [
+        OptimInfo(
+            name='cadafactor',
+            opt_class=Adafactor,
+            description='Cautious Adafactor',
+            defaults={'caution': True}
+        ),
+        OptimInfo(
+            name='cadafactorbv',
+            opt_class=AdafactorBigVision,
+            description='Cautious Big Vision Adafactor',
+            defaults={'caution': True}
+        ),
+        OptimInfo(
+            name='cadamw',
+            opt_class=AdamWLegacy,
+            description='Cautious AdamW',
+            has_betas=True,
+            defaults={'caution': True}
+        ),
+        OptimInfo(
+            name='cadopt',
+            opt_class=Adopt,
+            description='Cautious Adopt',
+            defaults={'caution': True}
+        ),
+        OptimInfo(
+            name='cadoptw',
+            opt_class=Adopt,
+            description='Cautious AdoptW (decoupled decay)',
+            defaults={'decoupled': True, 'caution': True}
+        ),
+        OptimInfo(
+            name='clamb',
+            opt_class=Lamb,
+            description='Cautious LAMB',
+            has_betas=True,
+            defaults={'caution': True}
+        ),
+        OptimInfo(
+            name='claprop',
+            opt_class=LaProp,
+            description='Cautious LaProp',
+            has_betas=True,
+            defaults={'caution': True}
+        ),
+        OptimInfo(
+            name='clion',
+            opt_class=Lion,
+            description='Cautious Lion',
+            has_eps=False,
+            has_betas=True,
+            defaults = {'caution': True}
+        ),
+        OptimInfo(
+            name='cmars',
+            opt_class=Mars,
+            description='Cautious MARS',
+            has_betas=True,
+            defaults={'caution': True}
+        ),
+        OptimInfo(
+            name='cnadamw',
+            opt_class=NAdamW,
+            description='Cautious NAdamW',
+            has_betas=True,
+            defaults={'caution': True}
+        ),
+        OptimInfo(
+            name='crmsproptf',
+            opt_class=RMSpropTF,
+            description='Cautious TensorFlow-style RMSprop',
+            has_momentum=True,
+            defaults={'alpha': 0.9, 'caution': True}
+        ),
+        OptimInfo(
+            name='csgdw',
+            opt_class=SGDW,
+            description='Cautious SGD with decoupled weight decay and Nesterov momentum',
+            has_eps=False,
+            has_momentum=True,
+            defaults={'nesterov': True, 'caution': True}
+        ),
+    ]
+    for opt in cautious_optimizers:
+        registry.register(opt)
+
+def _register_other_optimizers(registry: OptimizerRegistry) -> None:
+    """Register miscellaneous optimizers"""
+    other_optimizers = [
+        OptimInfo(
+            name='adabelief',
+            opt_class=AdaBelief,
+            description='Adapts learning rate based on gradient prediction error',
+            has_betas=True,
+            defaults={'rectify': False}
+        ),
+        OptimInfo(
+            name='radabelief',
+            opt_class=AdaBelief,
+            description='Rectified AdaBelief with variance adaptation',
+            has_betas=True,
+            defaults={'rectify': True}
+        ),
+        OptimInfo(
+            name='adadelta',
+            opt_class=torch.optim.Adadelta,
+            description='torch.optim.Adadelta, Adapts learning rates based on running windows of gradients'
+        ),
+        OptimInfo(
+            name='adagrad',
+            opt_class=torch.optim.Adagrad,
+            description='torch.optim.Adagrad, Adapts learning rates using cumulative squared gradients',
+            defaults={'eps': 1e-8}
+        ),
+        OptimInfo(
+            name='adan',
+            opt_class=Adan,
+            description='Adaptive Nesterov Momentum Algorithm',
+            defaults={'no_prox': False},
+            has_betas=True,
+            num_betas=3
+        ),
+        OptimInfo(
+            name='adanw',
+            opt_class=Adan,
+            description='Adaptive Nesterov Momentum with decoupled weight decay',
+            defaults={'no_prox': True},
+            has_betas=True,
+            num_betas=3
+        ),
+        OptimInfo(
+            name='adahessian',
+            opt_class=Adahessian,
+            description='An Adaptive Second Order Optimizer',
+            has_betas=True,
+            second_order=True,
+        ),
+        OptimInfo(
+            name='laprop',
+            opt_class=LaProp,
+            description='Separating Momentum and Adaptivity in Adam',
+            has_betas=True,
+        ),
+        OptimInfo(
+            name='lion',
+            opt_class=Lion,
+            description='Evolved Sign Momentum optimizer for improved convergence',
+            has_eps=False,
+            has_betas=True
+        ),
+        OptimInfo(
+            name='madgrad',
+            opt_class=MADGRAD,
+            description='Momentum-based Adaptive gradient method',
+            has_momentum=True
+        ),
+        OptimInfo(
+            name='madgradw',
+            opt_class=MADGRAD,
+            description='MADGRAD with decoupled weight decay',
+            has_momentum=True,
+            defaults={'decoupled_decay': True}
+        ),
+        OptimInfo(
+            name='mars',
+            opt_class=Mars,
+            description='Unleashing the Power of Variance Reduction for Training Large Models',
+            has_betas=True,
+        ),
+        OptimInfo(
+            name='novograd',
+            opt_class=NvNovoGrad,
+            description='Normalized Adam with L2 norm gradient normalization',
+            has_betas=True
+        ),
+        OptimInfo(
+            name='rmsprop',
+            opt_class=torch.optim.RMSprop,
+            description='torch.optim.RMSprop, Root Mean Square Propagation',
+            has_momentum=True,
+            defaults={'alpha': 0.9}
+        ),
+        OptimInfo(
+            name='rmsproptf',
+            opt_class=RMSpropTF,
+            description='TensorFlow-style RMSprop implementation, Root Mean Square Propagation',
+            has_momentum=True,
+            defaults={'alpha': 0.9}
+        ),
+    ]
+    for opt in other_optimizers:
+        registry.register(opt)
+    registry.register_foreach_default('lion')
+
+
+def _register_apex_optimizers(registry: OptimizerRegistry) -> None:
+    """Register APEX optimizers (lazy import)"""
+    apex_optimizers = [
+        OptimInfo(
+            name='fusedsgd',
+            opt_class='apex.optimizers.FusedSGD',
+            description='NVIDIA APEX fused SGD implementation for faster training',
+            has_eps=False,
+            has_momentum=True,
+            defaults={'nesterov': True}
+        ),
+        OptimInfo(
+            name='fusedadam',
+            opt_class='apex.optimizers.FusedAdam',
+            description='NVIDIA APEX fused Adam implementation',
+            has_betas=True,
+            defaults={'adam_w_mode': False}
+        ),
+        OptimInfo(
+            name='fusedadamw',
+            opt_class='apex.optimizers.FusedAdam',
+            description='NVIDIA APEX fused AdamW implementation',
+            has_betas=True,
+            defaults={'adam_w_mode': True}
+        ),
+        OptimInfo(
+            name='fusedlamb',
+            opt_class='apex.optimizers.FusedLAMB',
+            description='NVIDIA APEX fused LAMB implementation',
+            has_betas=True
+        ),
+        OptimInfo(
+            name='fusednovograd',
+            opt_class='apex.optimizers.FusedNovoGrad',
+            description='NVIDIA APEX fused NovoGrad implementation',
+            has_betas=True,
+            defaults={'betas': (0.95, 0.98)}
+        ),
+    ]
+    for opt in apex_optimizers:
+        registry.register(opt)
+
+
+def _register_bnb_optimizers(registry: OptimizerRegistry) -> None:
+    """Register bitsandbytes optimizers (lazy import)"""
+    bnb_optimizers = [
+        OptimInfo(
+            name='bnbsgd',
+            opt_class='bitsandbytes.optim.SGD',
+            description='bitsandbytes SGD',
+            has_eps=False,
+            has_momentum=True,
+            defaults={'nesterov': True}
+        ),
+        OptimInfo(
+            name='bnbsgd8bit',
+            opt_class='bitsandbytes.optim.SGD8bit',
+            description='bitsandbytes 8-bit SGD with dynamic quantization',
+            has_eps=False,
+            has_momentum=True,
+            defaults={'nesterov': True}
+        ),
+        OptimInfo(
+            name='bnbadam',
+            opt_class='bitsandbytes.optim.Adam',
+            description='bitsandbytes Adam',
+            has_betas=True
+        ),
+        OptimInfo(
+            name='bnbadam8bit',
+            opt_class='bitsandbytes.optim.Adam',
+            description='bitsandbytes 8-bit Adam with dynamic quantization',
+            has_betas=True
+        ),
+        OptimInfo(
+            name='bnbadamw',
+            opt_class='bitsandbytes.optim.AdamW',
+            description='bitsandbytes AdamW',
+            has_betas=True
+        ),
+        OptimInfo(
+            name='bnbadamw8bit',
+            opt_class='bitsandbytes.optim.AdamW',
+            description='bitsandbytes 8-bit AdamW with dynamic quantization',
+            has_betas=True
+        ),
+        OptimInfo(
+            'bnblion',
+            'bitsandbytes.optim.Lion',
+            description='bitsandbytes Lion',
+            has_eps=False,
+            has_betas=True
+        ),
+        OptimInfo(
+            'bnblion8bit',
+            'bitsandbytes.optim.Lion8bit',
+            description='bitsandbytes 8-bit Lion with dynamic quantization',
+            has_eps=False,
+            has_betas=True
+        ),
+        OptimInfo(
+            'bnbademamix',
+            'bitsandbytes.optim.AdEMAMix',
+            description='bitsandbytes AdEMAMix',
+            has_betas=True,
+            num_betas=3,
+        ),
+        OptimInfo(
+            'bnbademamix8bit',
+            'bitsandbytes.optim.AdEMAMix8bit',
+            description='bitsandbytes 8-bit AdEMAMix with dynamic quantization',
+            has_betas=True,
+            num_betas=3,
+        ),
+    ]
+    for opt in bnb_optimizers:
+        registry.register(opt)
+
+
+default_registry = OptimizerRegistry()
+
+def _register_default_optimizers() -> None:
+    """Register all default optimizers to the global registry."""
+    # Register all optimizer groups
+    _register_sgd_variants(default_registry)
+    _register_adam_variants(default_registry)
+    _register_lamb_lars(default_registry)
+    _register_other_optimizers(default_registry)
+    _register_apex_optimizers(default_registry)
+    _register_bnb_optimizers(default_registry)
+    _register_cautious_optimizers(default_registry)
+
+    # Register aliases
+    default_registry.register_alias('nesterov', 'sgd')
+    default_registry.register_alias('nesterovw', 'sgdw')
+
+
+# Initialize default registry
+_register_default_optimizers()
+
+# Public API
+
+def list_optimizers(
+        filter: Union[str, List[str]] = '',
+        exclude_filters: Optional[List[str]] = None,
+        with_description: bool = False,
+) -> List[Union[str, Tuple[str, str]]]:
+    """List available optimizer names, optionally filtered.
+
+    List all registered optimizers, with optional filtering using wildcard patterns.
+    Optimizers can be filtered using include and exclude patterns, and can optionally
+    return descriptions with each optimizer name.
+
+    Args:
+        filter: Wildcard style filter string or list of filter strings
+            (e.g., 'adam*' for all Adam variants, or ['adam*', '*8bit'] for
+            Adam variants and 8-bit optimizers). Empty string means no filtering.
+        exclude_filters: Optional list of wildcard patterns to exclude. For example,
+            ['*8bit', 'fused*'] would exclude 8-bit and fused implementations.
+        with_description: If True, returns tuples of (name, description) instead of
+            just names. Descriptions provide brief explanations of optimizer characteristics.
+
+    Returns:
+        If with_description is False:
+            List of optimizer names as strings (e.g., ['adam', 'adamw', ...])
+        If with_description is True:
+            List of tuples of (name, description) (e.g., [('adam', 'Adaptive Moment...'), ...])
+
+    Examples:
+        >>> list_optimizers()
+        ['adam', 'adamw', 'sgd', ...]
+
+        >>> list_optimizers(['la*', 'nla*'])  # List lamb & lars
+        ['lamb', 'lambc', 'larc', 'lars', 'nlarc', 'nlars']
+
+        >>> list_optimizers('*adam*', exclude_filters=['bnb*', 'fused*'])  # Exclude bnb & apex adam optimizers
+        ['adam', 'adamax', 'adamp', 'adamw', 'nadam', 'nadamw', 'radam']
+
+        >>> list_optimizers(with_description=True)  # Get descriptions
+        [('adabelief', 'Adapts learning rate based on gradient prediction error'),
+         ('adadelta', 'torch.optim Adadelta, Adapts learning rates based on running windows of gradients'),
+         ('adafactor', 'Memory-efficient implementation of Adam with factored gradients'),
+        ...]
+    """
+    return default_registry.list_optimizers(filter, exclude_filters, with_description)
+
+
+def get_optimizer_info(name: str) -> OptimInfo:
+    """Get the OptimInfo for an optimizer.
+
+    Args:
+        name: Name of the optimizer
+
+    Returns:
+        OptimInfo configuration
+
+    Raises:
+        ValueError: If optimizer is not found
+    """
+    return default_registry.get_optimizer_info(name)
+
+
+def get_optimizer_class(
+        name: str,
+        bind_defaults: bool = True,
+) -> Union[OptimType, OptimizerCallable]:
+    """Get optimizer class by name with option to bind default arguments.
+
+    Retrieves the optimizer class or a partial function with default arguments bound.
+    This allows direct instantiation of optimizers with their default configurations
+    without going through the full factory.
+
+    Args:
+        name: Name of the optimizer to retrieve (e.g., 'adam', 'sgd')
+        bind_defaults: If True, returns a partial function with default arguments from OptimInfo bound.
+            If False, returns the raw optimizer class.
+
+    Returns:
+        If bind_defaults is False:
+            The optimizer class (e.g., torch.optim.Adam)
+        If bind_defaults is True:
+            A partial function with default arguments bound
+
+    Raises:
+        ValueError: If optimizer name is not found in registry
+
+    Examples:
+        >>> # Get SGD with nesterov momentum default
+        >>> SGD = get_optimizer_class('sgd')  # nesterov=True bound
+        >>> opt = SGD(model.parameters(), lr=0.1, momentum=0.9)
+
+        >>> # Get raw optimizer class
+        >>> SGD = get_optimizer_class('sgd')
+        >>> opt = SGD(model.parameters(), lr=1e-3, momentum=0.9)
+
+    """
+    return default_registry.get_optimizer_class(name, bind_defaults=bind_defaults)
+
+
+def create_optimizer_v2(
+        model_or_params: Union[nn.Module, ParamsT],
+        opt: str = 'sgd',
+        lr: Optional[float] = None,
+        weight_decay: float = 0.,
+        momentum: float = 0.9,
+        foreach: Optional[bool] = None,
+        filter_bias_and_bn: bool = True,
+        layer_decay: Optional[float] = None,
+        param_group_fn: Optional[Callable[[nn.Module], ParamsT]] = None,
+        **kwargs: Any,
+) -> torch.optim.Optimizer:
+    """Create an optimizer instance via timm registry.
+
+    Creates and configures an optimizer with appropriate parameter groups and settings.
+    Supports automatic parameter group creation for weight decay and layer-wise learning
+    rates, as well as custom parameter grouping.
+
+    Args:
+        model_or_params: A PyTorch model or an iterable of parameters/parameter groups.
+            If a model is provided, parameters will be automatically extracted and grouped
+            based on the other arguments.
+        opt: Name of the optimizer to create (e.g., 'adam', 'adamw', 'sgd').
+            Use list_optimizers() to see available options.
+        lr: Learning rate. If None, will use the optimizer's default.
+        weight_decay: Weight decay factor. Will be used to create param groups if model_or_params is a model.
+        momentum: Momentum factor for optimizers that support it. Only used if the
+            chosen optimizer accepts a momentum parameter.
+        foreach: Enable/disable foreach (multi-tensor) implementation if available.
+            If None, will use optimizer-specific defaults.
+        filter_bias_and_bn: If True, bias, norm layer parameters (all 1d params) will not have
+            weight decay applied. Only used when model_or_params is a model and
+            weight_decay > 0.
+        layer_decay: Optional layer-wise learning rate decay factor. If provided,
+            learning rates will be scaled by layer_decay^(max_depth - layer_depth).
+            Only used when model_or_params is a model.
+        param_group_fn: Optional function to create custom parameter groups.
+            If provided, other parameter grouping options will be ignored.
+        **kwargs: Additional optimizer-specific arguments (e.g., betas for Adam).
+
+    Returns:
+        Configured optimizer instance.
+
+    Examples:
+        >>> # Basic usage with a model
+        >>> optimizer = create_optimizer_v2(model, 'adamw', lr=1e-3)
+
+        >>> # SGD with momentum and weight decay
+        >>> optimizer = create_optimizer_v2(
+        ...     model, 'sgd', lr=0.1, momentum=0.9, weight_decay=1e-4
+        ... )
+
+        >>> # Adam with layer-wise learning rate decay
+        >>> optimizer = create_optimizer_v2(
+        ...     model, 'adam', lr=1e-3, layer_decay=0.7
+        ... )
+
+        >>> # Custom parameter groups
+        >>> def group_fn(model):
+        ...     return [
+        ...         {'params': model.backbone.parameters(), 'lr': 1e-4},
+        ...         {'params': model.head.parameters(), 'lr': 1e-3}
+        ...     ]
+        >>> optimizer = create_optimizer_v2(
+        ...     model, 'sgd', param_group_fn=group_fn
+        ... )
+
+    Note:
+        Parameter group handling precedence:
+        1. If param_group_fn is provided, it will be used exclusively
+        2. If layer_decay is provided, layer-wise groups will be created
+        3. If weight_decay > 0 and filter_bias_and_bn is True, weight decay groups will be created
+        4. Otherwise, all parameters will be in a single group
+    """
+
+    return default_registry.create_optimizer(
+        model_or_params,
+        opt=opt,
+        lr=lr,
+        weight_decay=weight_decay,
+        momentum=momentum,
+        foreach=foreach,
+        weight_decay_exclude_1d=filter_bias_and_bn,
+        layer_decay=layer_decay,
+        param_group_fn=param_group_fn,
+        **kwargs
+    )
+
+
+def optimizer_kwargs(cfg):
+    """ cfg/argparse to kwargs helper
+    Convert optimizer args in argparse args or cfg like object to keyword args for updated create fn.
+    """
+    kwargs = dict(
+        opt=cfg.opt,
+        lr=cfg.lr,
+        weight_decay=cfg.weight_decay,
+        momentum=cfg.momentum,
+    )
+    if getattr(cfg, 'opt_eps', None) is not None:
+        kwargs['eps'] = cfg.opt_eps
+    if getattr(cfg, 'opt_betas', None) is not None:
+        kwargs['betas'] = cfg.opt_betas
+    if getattr(cfg, 'layer_decay', None) is not None:
+        kwargs['layer_decay'] = cfg.layer_decay
+    if getattr(cfg, 'opt_args', None) is not None:
+        kwargs.update(cfg.opt_args)
+    if getattr(cfg, 'opt_foreach', None) is not None:
+        kwargs['foreach'] = cfg.opt_foreach
+    return kwargs
+
+
+def create_optimizer(
+        args,
+        model: Union[nn.Module, ParamsT],
+        filter_bias_and_bn: bool = True,
+) -> torch.optim.Optimizer:
+    """ Legacy optimizer factory for backwards compatibility.
+    NOTE: Use create_optimizer_v2 for new code.
+    """
+    return create_optimizer_v2(
+        model,
+        **optimizer_kwargs(cfg=args),
+        filter_bias_and_bn=filter_bias_and_bn,
+    )
+

pytorch-image-models/timm/optim/_param_groups.py

@@ -0,0 +1,131 @@
+import logging
+from itertools import islice
+from typing import Collection, Optional
+
+from torch import nn as nn
+
+from timm.models import group_parameters
+
+
+_logger = logging.getLogger(__name__)
+
+
+def param_groups_weight_decay(
+        model: nn.Module,
+        weight_decay: float = 1e-5,
+        no_weight_decay_list: Collection[str] = (),
+):
+    no_weight_decay_list = set(no_weight_decay_list)
+    decay = []
+    no_decay = []
+    for name, param in model.named_parameters():
+        if not param.requires_grad:
+            continue
+
+        if param.ndim <= 1 or name.endswith(".bias") or name in no_weight_decay_list:
+            no_decay.append(param)
+        else:
+            decay.append(param)
+
+    return [
+        {'params': no_decay, 'weight_decay': 0.},
+        {'params': decay, 'weight_decay': weight_decay}]
+
+
+def _group(it, size):
+    it = iter(it)
+    return iter(lambda: tuple(islice(it, size)), ())
+
+
+def auto_group_layers(model, layers_per_group=12, num_groups=None):
+    def _in_head(n, hp):
+        if not hp:
+            return True
+        elif isinstance(hp, (tuple, list)):
+            return any([n.startswith(hpi) for hpi in hp])
+        else:
+            return n.startswith(hp)
+
+    head_prefix = getattr(model, 'pretrained_cfg', {}).get('classifier', None)
+    names_trunk = []
+    names_head = []
+    for n, _ in model.named_parameters():
+        names_head.append(n) if _in_head(n, head_prefix) else names_trunk.append(n)
+
+    # group non-head layers
+    num_trunk_layers = len(names_trunk)
+    if num_groups is not None:
+        layers_per_group = -(num_trunk_layers // -num_groups)
+    names_trunk = list(_group(names_trunk, layers_per_group))
+
+    num_trunk_groups = len(names_trunk)
+    layer_map = {n: i for i, l in enumerate(names_trunk) for n in l}
+    layer_map.update({n: num_trunk_groups for n in names_head})
+    return layer_map
+
+_layer_map = auto_group_layers  # backward compat
+
+
+def param_groups_layer_decay(
+        model: nn.Module,
+        weight_decay: float = 0.05,
+        no_weight_decay_list: Collection[str] = (),
+        weight_decay_exclude_1d: bool = True,
+        layer_decay: float = .75,
+        end_layer_decay: Optional[float] = None,
+        verbose: bool = False,
+):
+    """
+    Parameter groups for layer-wise lr decay & weight decay
+    Based on BEiT: https://github.com/microsoft/unilm/blob/master/beit/optim_factory.py#L58
+    """
+    no_weight_decay_list = set(no_weight_decay_list)
+    param_group_names = {}  # NOTE for debugging
+    param_groups = {}
+
+    if hasattr(model, 'group_matcher'):
+        # FIXME interface needs more work
+        layer_map = group_parameters(model, model.group_matcher(coarse=False), reverse=True)
+    else:
+        # fallback
+        layer_map = auto_group_layers(model)
+    num_layers = max(layer_map.values()) + 1
+    layer_max = num_layers - 1
+    layer_scales = list(layer_decay ** (layer_max - i) for i in range(num_layers))
+
+    for name, param in model.named_parameters():
+        if not param.requires_grad:
+            continue
+
+        # no decay: all 1D parameters and model specific ones
+        if (weight_decay_exclude_1d and param.ndim <= 1) or name in no_weight_decay_list:
+            g_decay = "no_decay"
+            this_decay = 0.
+        else:
+            g_decay = "decay"
+            this_decay = weight_decay
+
+        layer_id = layer_map.get(name, layer_max)
+        group_name = "layer_%d_%s" % (layer_id, g_decay)
+
+        if group_name not in param_groups:
+            this_scale = layer_scales[layer_id]
+            param_group_names[group_name] = {
+                "lr_scale": this_scale,
+                "weight_decay": this_decay,
+                "param_names": [],
+            }
+            param_groups[group_name] = {
+                "lr_scale": this_scale,
+                "weight_decay": this_decay,
+                "params": [],
+            }
+
+        param_group_names[group_name]["param_names"].append(name)
+        param_groups[group_name]["params"].append(param)
+
+    if verbose:
+        import json
+        _logger.info("parameter groups: \n%s" % json.dumps(param_group_names, indent=2))
+
+    return list(param_groups.values())

pytorch-image-models/timm/optim/adabelief.py

@@ -0,0 +1,218 @@
+import math
+import torch
+from torch.optim.optimizer import Optimizer
+
+
+class AdaBelief(Optimizer):
+    r"""Implements AdaBelief algorithm. Modified from Adam in PyTorch
+
+    Arguments:
+        params (iterable): iterable of parameters to optimize or dicts defining
+            parameter groups
+        lr (float, optional): learning rate (default: 1e-3)
+        betas (Tuple[float, float], optional): coefficients used for computing
+            running averages of gradient and its square (default: (0.9, 0.999))
+        eps (float, optional): term added to the denominator to improve
+            numerical stability (default: 1e-16)
+        weight_decay (float, optional): weight decay (L2 penalty) (default: 0)
+        amsgrad (boolean, optional): whether to use the AMSGrad variant of this
+            algorithm from the paper `On the Convergence of Adam and Beyond`_
+            (default: False)
+        decoupled_decay (boolean, optional): (default: True) If set as True, then
+            the optimizer uses decoupled weight decay as in AdamW
+        fixed_decay (boolean, optional): (default: False) This is used when weight_decouple
+            is set as True.
+            When fixed_decay == True, the weight decay is performed as
+            $W_{new} = W_{old} - W_{old} \times decay$.
+            When fixed_decay == False, the weight decay is performed as
+            $W_{new} = W_{old} - W_{old} \times decay \times lr$. Note that in this case, the
+            weight decay ratio decreases with learning rate (lr).
+        rectify (boolean, optional): (default: True) If set as True, then perform the rectified
+            update similar to RAdam
+        degenerated_to_sgd (boolean, optional) (default:True) If set as True, then perform SGD update
+            when variance of gradient is high
+    reference: AdaBelief Optimizer, adapting stepsizes by the belief in observed gradients, NeurIPS 2020
+
+    For a complete table of recommended hyperparameters, see https://github.com/juntang-zhuang/Adabelief-Optimizer'
+    For example train/args for EfficientNet see these gists
+      - link to train_scipt: https://gist.github.com/juntang-zhuang/0a501dd51c02278d952cf159bc233037
+      - link to args.yaml: https://gist.github.com/juntang-zhuang/517ce3c27022b908bb93f78e4f786dc3
+    """
+
+    def __init__(
+            self,
+            params,
+            lr=1e-3,
+            betas=(0.9, 0.999),
+            eps=1e-16,
+            weight_decay=0,
+            amsgrad=False,
+            decoupled_decay=True,
+            fixed_decay=False,
+            rectify=True,
+            degenerated_to_sgd=True,
+    ):
+        if not 0.0 <= lr:
+            raise ValueError("Invalid learning rate: {}".format(lr))
+        if not 0.0 <= eps:
+            raise ValueError("Invalid epsilon value: {}".format(eps))
+        if not 0.0 <= betas[0] < 1.0:
+            raise ValueError("Invalid beta parameter at index 0: {}".format(betas[0]))
+        if not 0.0 <= betas[1] < 1.0:
+            raise ValueError("Invalid beta parameter at index 1: {}".format(betas[1]))
+
+        if isinstance(params, (list, tuple)) and len(params) > 0 and isinstance(params[0], dict):
+            for param in params:
+                if 'betas' in param and (param['betas'][0] != betas[0] or param['betas'][1] != betas[1]):
+                    param['buffer'] = [[None, None, None] for _ in range(10)]
+
+        defaults = dict(
+            lr=lr,
+            betas=betas,
+            eps=eps,
+            weight_decay=weight_decay,
+            amsgrad=amsgrad,
+            degenerated_to_sgd=degenerated_to_sgd,
+            decoupled_decay=decoupled_decay,
+            rectify=rectify,
+            fixed_decay=fixed_decay,
+            buffer=[[None, None, None] for _ in range(10)]
+        )
+        super(AdaBelief, self).__init__(params, defaults)
+
+    def __setstate__(self, state):
+        super(AdaBelief, self).__setstate__(state)
+        for group in self.param_groups:
+            group.setdefault('amsgrad', False)
+
+    @torch.no_grad()
+    def reset(self):
+        for group in self.param_groups:
+            for p in group['params']:
+                state = self.state[p]
+                amsgrad = group['amsgrad']
+
+                # State initialization
+                state['step'] = 0
+                # Exponential moving average of gradient values
+                state['exp_avg'] = torch.zeros_like(p)
+
+                # Exponential moving average of squared gradient values
+                state['exp_avg_var'] = torch.zeros_like(p)
+                if amsgrad:
+                    # Maintains max of all exp. moving avg. of sq. grad. values
+                    state['max_exp_avg_var'] = torch.zeros_like(p)
+
+    @torch.no_grad()
+    def step(self, closure=None):
+        """Performs a single optimization step.
+        Arguments:
+            closure (callable, optional): A closure that reevaluates the model
+                and returns the loss.
+        """
+        loss = None
+        if closure is not None:
+            with torch.enable_grad():
+                loss = closure()
+
+        for group in self.param_groups:
+            for p in group['params']:
+                if p.grad is None:
+                    continue
+                grad = p.grad
+                if grad.dtype in {torch.float16, torch.bfloat16}:
+                    grad = grad.float()
+                if grad.is_sparse:
+                    raise RuntimeError(
+                        'AdaBelief does not support sparse gradients, please consider SparseAdam instead')
+
+                p_fp32 = p
+                if p.dtype in {torch.float16, torch.bfloat16}:
+                    p_fp32 = p_fp32.float()
+
+                amsgrad = group['amsgrad']
+                beta1, beta2 = group['betas']
+                state = self.state[p]
+                # State initialization
+                if len(state) == 0:
+                    state['step'] = 0
+                    # Exponential moving average of gradient values
+                    state['exp_avg'] = torch.zeros_like(p_fp32)
+                    # Exponential moving average of squared gradient values
+                    state['exp_avg_var'] = torch.zeros_like(p_fp32)
+                    if amsgrad:
+                        # Maintains max of all exp. moving avg. of sq. grad. values
+                        state['max_exp_avg_var'] = torch.zeros_like(p_fp32)
+                
+                # perform weight decay, check if decoupled weight decay
+                if group['decoupled_decay']:
+                    if not group['fixed_decay']:
+                        p_fp32.mul_(1.0 - group['lr'] * group['weight_decay'])
+                    else:
+                        p_fp32.mul_(1.0 - group['weight_decay'])
+                else:
+                    if group['weight_decay'] != 0:
+                        grad.add_(p_fp32, alpha=group['weight_decay'])
+
+                # get current state variable
+                exp_avg, exp_avg_var = state['exp_avg'], state['exp_avg_var']
+
+                state['step'] += 1
+                bias_correction1 = 1 - beta1 ** state['step']
+                bias_correction2 = 1 - beta2 ** state['step']
+
+                # Update first and second moment running average
+                exp_avg.mul_(beta1).add_(grad, alpha=1 - beta1)
+                grad_residual = grad - exp_avg
+                exp_avg_var.mul_(beta2).addcmul_(grad_residual, grad_residual, value=1 - beta2)
+
+                if amsgrad:
+                    max_exp_avg_var = state['max_exp_avg_var']
+                    # Maintains the maximum of all 2nd moment running avg. till now
+                    torch.max(max_exp_avg_var, exp_avg_var.add_(group['eps']), out=max_exp_avg_var)
+
+                    # Use the max. for normalizing running avg. of gradient
+                    denom = (max_exp_avg_var.sqrt() / math.sqrt(bias_correction2)).add_(group['eps'])
+                else:
+                    denom = (exp_avg_var.add_(group['eps']).sqrt() / math.sqrt(bias_correction2)).add_(group['eps'])
+                
+                # update
+                if not group['rectify']:
+                    # Default update
+                    step_size = group['lr'] / bias_correction1
+                    p_fp32.addcdiv_(exp_avg, denom, value=-step_size)
+                else:
+                    # Rectified update, forked from RAdam
+                    buffered = group['buffer'][int(state['step'] % 10)]
+                    if state['step'] == buffered[0]:
+                        num_sma, step_size = buffered[1], buffered[2]
+                    else:
+                        buffered[0] = state['step']
+                        beta2_t = beta2 ** state['step']
+                        num_sma_max = 2 / (1 - beta2) - 1
+                        num_sma = num_sma_max - 2 * state['step'] * beta2_t / (1 - beta2_t)
+                        buffered[1] = num_sma
+
+                        # more conservative since it's an approximated value
+                        if num_sma >= 5:
+                            step_size = math.sqrt(
+                                (1 - beta2_t) *
+                                (num_sma - 4) / (num_sma_max - 4) *
+                                (num_sma - 2) / num_sma *
+                                num_sma_max / (num_sma_max - 2)) / (1 - beta1 ** state['step'])
+                        elif group['degenerated_to_sgd']:
+                            step_size = 1.0 / (1 - beta1 ** state['step'])
+                        else:
+                            step_size = -1
+                        buffered[2] = step_size
+
+                    if num_sma >= 5:
+                        denom = exp_avg_var.sqrt().add_(group['eps'])
+                        p_fp32.addcdiv_(exp_avg, denom, value=-step_size * group['lr'])
+                    elif step_size > 0:
+                        p_fp32.add_(exp_avg, alpha=-step_size * group['lr'])
+                
+                if p.dtype in {torch.float16, torch.bfloat16}:
+                    p.copy_(p_fp32)
+
+        return loss

pytorch-image-models/timm/optim/adafactor.py

@@ -0,0 +1,229 @@
+""" Adafactor Optimizer
+
+Lifted from https://github.com/pytorch/fairseq/blob/master/fairseq/optim/adafactor.py
+
+Modified by Ross Wightman to fix some issues with factorization dims for non nn.Linear layers
+
+Original header/copyright below.
+"""
+# Copyright (c) Facebook, Inc. and its affiliates.
+#
+# This source code is licensed under the MIT license found in the
+# LICENSE file in the root directory of this source tree.
+import math
+from typing import Optional, Tuple
+
+import torch
+
+from ._types import ParamsT
+
+
+class Adafactor(torch.optim.Optimizer):
+    """Implements Adafactor algorithm.
+
+    This implementation is based on: `Adafactor: Adaptive Learning Rates with Sublinear Memory Cost`
+    (see https://arxiv.org/abs/1804.04235)
+
+    Note that this optimizer internally adjusts the learning rate depending on the
+    *scale_parameter*, *relative_step* and *warmup_init* options.
+
+    To use a manual (external) learning rate schedule you should set `scale_parameter=False` and
+    `relative_step=False`.
+
+    Ags:
+        params: iterable of parameters to optimize or dicts defining parameter groups
+        lr: external learning rate
+        eps: regularization constants for square gradient and parameter scale respectively
+        eps_scale: regularization constants for parameter scale respectively
+        clip_threshold: threshold of root-mean-square of final gradient update
+        decay_rate: coefficient used to compute running averages of square gradient
+        beta1: coefficient used for computing running averages of gradient
+        weight_decay: weight decay
+        scale_parameter: if True, learning rate is scaled by root-mean-square of parameter
+        warmup_init: time-dependent learning rate computation depends on whether warm-up initialization is being used
+    """
+
+    def __init__(
+            self,
+            params: ParamsT,
+            lr: Optional[float] = None,
+            eps: float = 1e-30,
+            eps_scale: float = 1e-3,
+            clip_threshold: float = 1.0,
+            decay_rate: float = -0.8,
+            betas: Optional[Tuple[float, float]] = None,
+            weight_decay: float = 0.0,
+            scale_parameter: bool = True,
+            warmup_init: bool = False,
+            min_dim_size_to_factor: int = 16,
+            caution: bool = False,
+    ):
+        relative_step = not lr
+        if warmup_init and not relative_step:
+            raise ValueError('warmup_init requires relative_step=True')
+
+        beta1 = None if betas is None else betas[0]   # make it compat with standard betas arg
+        defaults = dict(
+            lr=lr,
+            eps=eps,
+            eps_scale=eps_scale,
+            clip_threshold=clip_threshold,
+            decay_rate=decay_rate,
+            beta1=beta1,
+            weight_decay=weight_decay,
+            scale_parameter=scale_parameter,
+            relative_step=relative_step,
+            warmup_init=warmup_init,
+            min_dim_size_to_factor=min_dim_size_to_factor,
+            caution=caution,
+        )
+        super(Adafactor, self).__init__(params, defaults)
+
+    def __setstate__(self, state):
+        super().__setstate__(state)
+        for group in self.param_groups:
+            group.setdefault('caution', False)
+            group.setdefault('min_dim_size_to_factor', 16)
+
+    @staticmethod
+    def _get_lr(param_group, param_state):
+        if param_group['relative_step']:
+            min_step = 1e-6 * param_state['step'] if param_group['warmup_init'] else 1e-2
+            lr_t = min(min_step, 1.0 / math.sqrt(param_state['step']))
+            param_scale = 1.0
+            if param_group['scale_parameter']:
+                param_scale = max(param_group['eps_scale'], param_state['RMS'])
+            param_group['lr'] = lr_t * param_scale
+        return param_group['lr']
+
+    @staticmethod
+    def _get_options(param_group, param_shape, min_size_to_factor=16):
+        use_first_moment = param_group['beta1'] is not None
+        factored = None
+        ndim = len(param_shape)
+        # Use a simple heuristic to pick factorization row & col, note other PyTorch impl tend to
+        # always use -2, -1 BUT this will not pick correct dims for convolutions. This is a simple
+        # approach that should work in most cases, compare to the slightly more involved approach
+        # in AdafactorBigVision that sorts dims by size, please report if wrong dims chosen.
+        if ndim > 2 and param_shape[0] > min_size_to_factor and param_shape[1] > min_size_to_factor:
+            # nD convs in torch are ND + 2 dim weights with leading in/out chs
+            factored = 0, 1
+        elif ndim >= 2 and param_shape[-2] > min_size_to_factor and param_shape[-1] > min_size_to_factor:
+            # if the criteria above didn't match, test trailing dims for eligibility as per original impl
+            factored = ndim - 2, ndim - 1
+
+        return factored, use_first_moment
+
+    @staticmethod
+    def _rms(tensor):
+        return tensor.norm(2) / (tensor.numel() ** 0.5)
+
+    def _approx_sq_grad(self, exp_avg_sq_row, exp_avg_sq_col, dim_col, dim_row):
+        # from our dim heuristic, always dim_col < dim_row, so col reduction dim for factored row = dim_col
+        r_factor = (exp_avg_sq_row / exp_avg_sq_row.mean(dim=dim_col, keepdim=True)).rsqrt_().unsqueeze(dim_row)
+        c_factor = exp_avg_sq_col.unsqueeze(dim_col).rsqrt()
+        return torch.mul(r_factor, c_factor)
+
+    @torch.no_grad()
+    def step(self, closure=None):
+        """Performs a single optimization step.
+        Arguments:
+            closure (callable, optional): A closure that reevaluates the model and returns the loss.
+        """
+        loss = None
+        if closure is not None:
+            with torch.enable_grad():
+                loss = closure()
+
+        for group in self.param_groups:
+            for p in group['params']:
+                if p.grad is None:
+                    continue
+                grad = p.grad
+                if grad.dtype in {torch.float16, torch.bfloat16}:
+                    grad = grad.float()
+                if grad.is_sparse:
+                    raise RuntimeError('Adafactor does not support sparse gradients.')
+
+                state = self.state[p]
+
+                factored_dims, use_first_moment = self._get_options(
+                    group,
+                    grad.shape,
+                    min_size_to_factor=group['min_dim_size_to_factor'],
+                )
+                # State Initialization
+                if len(state) == 0:
+                    state['step'] = 0
+
+                    if use_first_moment:
+                        # Exponential moving average of gradient values
+                        state['exp_avg'] = torch.zeros_like(grad)
+                    if factored_dims is not None:
+                        dim_col, dim_row = factored_dims
+                        def _remove_dim(shape, dim):
+                            return shape[:dim] + shape[dim + 1:]
+                        state['exp_avg_sq_row'] = torch.zeros(_remove_dim(grad.shape, dim_row)).to(grad)
+                        state['exp_avg_sq_col'] = torch.zeros(_remove_dim(grad.shape, dim_col)).to(grad)
+                    else:
+                        state['exp_avg_sq'] = torch.zeros_like(grad)
+
+                    state['RMS'] = 0
+                else:
+                    if use_first_moment:
+                        state['exp_avg'] = state['exp_avg'].to(grad)
+                    if factored_dims is not None:
+                        state['exp_avg_sq_row'] = state['exp_avg_sq_row'].to(grad)
+                        state['exp_avg_sq_col'] = state['exp_avg_sq_col'].to(grad)
+                    else:
+                        state['exp_avg_sq'] = state['exp_avg_sq'].to(grad)
+
+                p_fp32 = p
+                if p.dtype in {torch.float16, torch.bfloat16}:
+                    p_fp32 = p_fp32.float()
+
+                state['step'] += 1
+                state['RMS'] = self._rms(p_fp32)
+                lr_t = self._get_lr(group, state)
+
+                beta2t = 1.0 - math.pow(state['step'], group['decay_rate'])
+                update = grad ** 2 + group['eps']
+                if factored_dims is not None:
+                    dim_col, dim_row = factored_dims
+                    exp_avg_sq_row = state['exp_avg_sq_row']
+                    exp_avg_sq_col = state['exp_avg_sq_col']
+
+                    exp_avg_sq_row.mul_(beta2t).add_(update.mean(dim=dim_row), alpha=1.0 - beta2t)
+                    exp_avg_sq_col.mul_(beta2t).add_(update.mean(dim=dim_col), alpha=1.0 - beta2t)
+
+                    # Approximation of exponential moving average of square of gradient
+                    update = self._approx_sq_grad(exp_avg_sq_row, exp_avg_sq_col, dim_col, dim_row)
+                    update.mul_(grad)
+                else:
+                    exp_avg_sq = state['exp_avg_sq']
+
+                    exp_avg_sq.mul_(beta2t).add_(update, alpha=1.0 - beta2t)
+                    update = exp_avg_sq.rsqrt().mul_(grad)
+
+                update.div_((self._rms(update) / group['clip_threshold']).clamp_(min=1.0))
+                update.mul_(lr_t)
+
+                if use_first_moment:
+                    exp_avg = state['exp_avg']
+                    exp_avg.mul_(group['beta1']).add_(update, alpha=1 - group['beta1'])
+                    if group['caution']:
+                        # Apply caution as per 'Cautious Optimizers' - https://arxiv.org/abs/2411.16085
+                        mask = (exp_avg * grad > 0).to(grad.dtype)
+                        mask.div_(mask.mean().clamp_(min=1e-3))
+                        update = exp_avg * mask
+                    else:
+                        update = exp_avg
+
+                if group['weight_decay'] != 0:
+                    p_fp32.add_(p_fp32, alpha=-group['weight_decay'] * lr_t)
+
+                p_fp32.add_(-update)
+                if p.dtype in {torch.float16, torch.bfloat16}:
+                    p.copy_(p_fp32)
+
+        return loss