""" PyTorch Lamb optimizer w/ behaviour similar to NVIDIA FusedLamb This optimizer code was adapted from the following (starting with latest) * https://github.com/HabanaAI/Model-References/blob/2b435114fe8e31f159b1d3063b8280ae37af7423/PyTorch/nlp/bert/pretraining/lamb.py * https://github.com/NVIDIA/DeepLearningExamples/blob/master/PyTorch/LanguageModeling/Transformer-XL/pytorch/lamb.py * https://github.com/cybertronai/pytorch-lamb Use FusedLamb if you can (GPU). The reason for including this variant of Lamb is to have a version that is similar in behaviour to APEX FusedLamb if you aren't using NVIDIA GPUs or cannot install/use APEX. In addition to some cleanup, this Lamb impl has been modified to support PyTorch XLA and has been tested on TPU. References for added functionality: Cautious Optimizers: https://arxiv.org/abs/2411.16085 Why Gradients Rapidly Increase Near the End of Training: https://arxiv.org/abs/2506.02285 Original copyrights for above sources are below. Modifications Copyright 2021 Ross Wightman """ # Copyright (c) 2021, Habana Labs Ltd. All rights reserved. # Copyright (c) 2019-2020, NVIDIA CORPORATION. All rights reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # MIT License # # Copyright (c) 2019 cybertronai # # Permission is hereby granted, free of charge, to any person obtaining a copy # of this software and associated documentation files (the "Software"), to deal # in the Software without restriction, including without limitation the rights # to use, copy, modify, merge, publish, distribute, sublicense, and/or sell # copies of the Software, and to permit persons to whom the Software is # furnished to do so, subject to the following conditions: # # The above copyright notice and this permission notice shall be included in all # copies or substantial portions of the Software. # # THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR # IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, # FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE # AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER # LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, # OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE # SOFTWARE. import math from typing import Optional, Tuple import torch from torch.optim import Optimizer from ._types import ParamsT class Lamb(Optimizer): """Implements a pure pytorch variant of FuseLAMB (NvLamb variant) optimizer from apex.optimizers.FusedLAMB reference: https://github.com/NVIDIA/DeepLearningExamples/blob/master/PyTorch/LanguageModeling/Transformer-XL/pytorch/lamb.py LAMB was proposed in: - Large Batch Optimization for Deep Learning - Training BERT in 76 minutes: https://arxiv.org/abs/1904.00962 - On the Convergence of Adam and Beyond: https://openreview.net/forum?id=ryQu7f-RZ Args: params: Iterable of parameters to optimize or dicts defining parameter groups. lr: Learning rate betas: Coefficients used for computing running averages of gradient and its norm. eps: Term added to the denominator to improve numerical stability. weight_decay: Weight decay grad_averaging: Whether apply (1-beta2) to grad when calculating running averages of gradient. max_grad_norm: Value used to clip global grad norm. trust_clip: Enable LAMBC trust ratio clipping. always_adapt: Apply adaptive learning rate to 0.0 weight decay parameter. caution: Apply caution. decoupled: apply decoupled weight decay corrected_weight_decay: apply corrected weight decay (lr**2 / max_lr) when using decoupled_decay """ def __init__( self, params: ParamsT, lr: float = 1e-3, bias_correction: bool = True, betas: Tuple[float, float] = (0.9, 0.999), eps: float = 1e-6, weight_decay: float = 0.01, grad_averaging: bool = True, max_grad_norm: Optional[float] = 1.0, trust_clip: bool = False, always_adapt: bool = False, caution: bool = False, decoupled_decay: bool = False, corrected_weight_decay: bool = False, ): defaults = dict( lr=lr, bias_correction=bias_correction, betas=betas, eps=eps, weight_decay=weight_decay, grad_averaging=grad_averaging, max_grad_norm=max_grad_norm, trust_clip=trust_clip, always_adapt=always_adapt, caution=caution, decoupled_decay=decoupled_decay, corrected_weight_decay=corrected_weight_decay, ) super().__init__(params, defaults) def __setstate__(self, state): super().__setstate__(state) for group in self.param_groups: group.setdefault('caution', False) group.setdefault('decoupled_decay', False) group.setdefault('corrected_weight_decay', False) def _get_clip_grad_norm(self): max_grad_norm = self.defaults['max_grad_norm'] if max_grad_norm is None: return None norms = [] for group in self.param_groups: for p in group['params']: if p.grad is None: continue grad = p.grad if grad.is_sparse: raise RuntimeError('Lamb does not support sparse gradients, consider SparseAdam instead.') norms.append(torch.linalg.vector_norm(grad)) global_norm = torch.linalg.vector_norm(torch.stack(norms)) clip_global_norm = (global_norm / max_grad_norm).clamp_(min=1.0) return clip_global_norm @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() clip_grad_norm = self._get_clip_grad_norm() # None if disabled for group in self.param_groups: bias_correction = 1 if group['bias_correction'] else 0 beta1, beta2 = group['betas'] grad_averaging = 1 if group['grad_averaging'] else 0 beta3 = 1 - beta1 if grad_averaging else 1.0 # assume same step across group now to simplify things # per parameter step can be easily support by making it tensor, or pass list into kernel if 'step' in group: group['step'] += 1 else: group['step'] = 1 if bias_correction: bias_correction1 = 1 - beta1 ** group['step'] bias_correction2 = 1 - beta2 ** group['step'] else: bias_correction1, bias_correction2 = 1.0, 1.0 for p in group['params']: if p.grad is None: continue grad = p.grad if clip_grad_norm is not None: grad.div_(clip_grad_norm) state = self.state[p] # State initialization if len(state) == 0: # Exponential moving average of gradient valuesa state['exp_avg'] = torch.zeros_like(p) # Exponential moving average of squared gradient values state['exp_avg_sq'] = torch.zeros_like(p) exp_avg, exp_avg_sq = state['exp_avg'], state['exp_avg_sq'] # Decay the first and second moment running average coefficient exp_avg.mul_(beta1).add_(grad, alpha=beta3) # m_t exp_avg_sq.mul_(beta2).addcmul_(grad, grad, value=1 - beta2) # v_t denom = (exp_avg_sq.sqrt() / math.sqrt(bias_correction2)).add_(group['eps']) update = (exp_avg / bias_correction1).div_(denom) if group['caution']: # Apply caution as per 'Cautious Optimizers' - https://arxiv.org/abs/2411.16085 mask = (update * grad > 0).to(grad.dtype) mask.div_(mask.mean().clamp_(min=1e-3)) update.mul_(mask) weight_decay = group['weight_decay'] if weight_decay != 0: if group.get('decoupled_decay', False): if group['corrected_weight_decay']: wd_scale = group['lr'] ** 2 / self.defaults['lr'] else: wd_scale = group['lr'] p.add_(p, alpha=-wd_scale * weight_decay) else: update.add_(p, alpha=weight_decay) if weight_decay != 0 or group['always_adapt']: # Layer-wise LR adaptation. By default, skip adaptation on parameters that are # excluded from weight decay, unless always_adapt == True, then always enabled. w_norm = p.norm(2.0) g_norm = update.norm(2.0) trust_ratio = w_norm / g_norm # FIXME nested where required since logical and/or not working in PT XLA # Set the ratio to 1.0 (no change) if either weight norm or grad norm is zero trust_ratio = torch.where( w_norm > 0, torch.where(g_norm > 0, trust_ratio, 1.0), 1.0, ) if group['trust_clip']: # LAMBC trust clipping, upper bound fixed at one trust_ratio = torch.clamp(trust_ratio, max=1.0) update.mul_(trust_ratio) p.add_(update, alpha=-group['lr']) return loss