feature: implement AdaGC optimizer

Author: kozistr

pytorch_optimizer/optimizer/adagc.py

@@ -0,0 +1,139 @@
+import math
+
+import torch
+
+from pytorch_optimizer.base.exception import NoSparseGradientError
+from pytorch_optimizer.base.optimizer import BaseOptimizer
+from pytorch_optimizer.base.type import BETAS, CLOSURE, DEFAULTS, LOSS, PARAMETERS
+from pytorch_optimizer.optimizer.utils import get_global_gradient_norm
+
+
+class AdaGC(BaseOptimizer):
+    r"""Improving Training Stability for Large Language Model Pretraining.
+
+    :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 beta: float. smoothing coefficient for EMA.
+    :param lambda_abs: float. absolute clipping threshold to prevent unstable updates from gradient explosions.
+    :param lambda_rel: float. relative clipping threshold to prevent unstable updates from gradient explosions.
+    :param warmup_steps: int. warmup steps.
+    :param weight_decay: float. weight decay (L2 penalty).
+    :param weight_decouple: bool. the optimizer uses decoupled weight decay as in AdamW.
+    :param fixed_decay: bool. fix weight decay.
+    :param eps: float. term added to the denominator to improve numerical stability.
+    """
+
+    def __init__(
+        self,
+        params: PARAMETERS,
+        lr: float = 1e-3,
+        betas: BETAS = (0.9, 0.999),
+        beta: float = 0.98,
+        lambda_abs: float = 1.0,
+        lambda_rel: float = 1.05,
+        warmup_steps: int = 100,
+        weight_decay: float = 1e-1,
+        weight_decouple: bool = True,
+        fixed_decay: bool = False,
+        eps: float = 1e-8,
+        **kwargs,
+    ):
+        self.validate_learning_rate(lr)
+        self.validate_betas(betas)
+        self.validate_range(beta, 'beta', 0.0, 1.0, '[)')
+        self.validate_positive(lambda_abs, 'lambda_abs')
+        self.validate_positive(lambda_rel, 'lambda_rel')
+        self.validate_non_negative(warmup_steps, 'warmup_steps')
+        self.validate_non_negative(weight_decay, 'weight_decay')
+        self.validate_non_negative(eps, 'eps')
+
+        defaults: DEFAULTS = {
+            'lr': lr,
+            'betas': betas,
+            'beta': beta,
+            'lambda_abs': lambda_abs,
+            'lambda_rel': lambda_rel,
+            'warmup_steps': warmup_steps,
+            'weight_decay': weight_decay,
+            'weight_decouple': weight_decouple,
+            'fixed_decay': fixed_decay,
+            'eps': eps,
+        }
+        super().__init__(params, defaults)
+
+    def __str__(self) -> str:
+        return 'AdaGC'
+
+    @torch.no_grad()
+    def reset(self):
+        pass
+
+    @torch.no_grad()
+    def step(self, closure: CLOSURE = None) -> LOSS:
+        loss: LOSS = None
+        if closure is not None:
+            with torch.enable_grad():
+                loss = closure()
+
+        for group in self.param_groups:
+            if 'step' in group:
+                group['step'] += 1
+            else:
+                group['step'] = 1
+
+            beta1, beta2 = group['betas']
+
+            bias_correction1: float = self.debias(beta1, group['step'])
+            bias_correction2_sq: float = math.sqrt(self.debias(beta2, group['step']))
+
+            for p in group['params']:
+                if p.grad is None:
+                    continue
+
+                grad = p.grad
+                if grad.is_sparse:
+                    raise NoSparseGradientError(str(self))
+
+                state = self.state[p]
+
+                if 'exp_avg' not in state:
+                    state['exp_avg'] = torch.zeros_like(grad)
+                    state['exp_avg_sq'] = torch.zeros_like(grad)
+                    state['gamma'] = torch.empty((1,), device=grad.device, dtype=grad.dtype)
+
+                self.apply_weight_decay(
+                    p=p,
+                    grad=grad,
+                    lr=group['lr'],
+                    weight_decay=group['weight_decay'],
+                    weight_decouple=group['weight_decouple'],
+                    fixed_decay=group['fixed_decay'],
+                )
+
+                gamma = state['gamma']
+
+                if group['step'] < group['warmup_steps']:
+                    grad_norm = get_global_gradient_norm(self.param_groups).add_(group['eps'])
+
+                    h_t = min(group['lambda_abs'] / grad_norm, 1.0)
+                    g_hat = grad.mul(h_t)
+
+                    g_hat_norm = g_hat.norm()
+
+                    gamma.copy_(g_hat_norm if group['step'] == 1 else min(gamma, g_hat_norm))
+                else:
+                    h_t = min(group['lambda_rel'] * gamma / grad.norm(), 1.0)
+                    g_hat = grad.mul(h_t)
+
+                    gamma.mul_(group['beta']).add_(g_hat.norm(), alpha=1.0 - group['beta'])
+
+                exp_avg, exp_avg_sq = state['exp_avg'], state['exp_avg_sq']
+                exp_avg.mul_(beta1).add_(g_hat, alpha=1.0 - beta1)
+                exp_avg_sq.mul_(beta2).addcmul_(g_hat, g_hat, value=1.0 - beta2)
+
+                update = (exp_avg / bias_correction1) / exp_avg_sq.sqrt().div_(bias_correction2_sq).add_(group['eps'])
+
+                p.add_(update, alpha=-group['lr'])
+
+        return loss