Skip to contents

Adafactor is commonly used in NLP tasks, and has the advantage of taking less memory because it only saves partial information of previous gradients.

The default argument setup is based on the original paper (see reference). When gradients are of dimension > 2, Adafactor optimizer will delete the last 2 dimensions separately in its accumulator variables.


  learning_rate = 0.001,
  beta_2_decay = -0.8,
  epsilon_1 = 1e-30,
  epsilon_2 = 0.001,
  clip_threshold = 1,
  relative_step = TRUE,
  weight_decay = NULL,
  clipnorm = NULL,
  clipvalue = NULL,
  global_clipnorm = NULL,
  use_ema = FALSE,
  ema_momentum = 0.99,
  ema_overwrite_frequency = NULL,
  name = "adafactor",
  loss_scale_factor = NULL,
  gradient_accumulation_steps = NULL



A float, a LearningRateSchedule() instance, or a callable that takes no arguments and returns the actual value to use. The learning rate. Defaults to 0.001.


float, defaults to -0.8. The decay rate of beta_2.


float, defaults to 1e-30. A small offset to keep denominator away from 0.


float, defaults to 1e-3. A small offset to avoid learning rate becoming too small by time.


float, defaults to 1.0. Clipping threshold. This is a part of Adafactor algorithm, independent from clipnorm, clipvalue, and global_clipnorm.


bool, defaults to TRUE. If learning_rate is a constant and relative_step=TRUE, learning rate will be adjusted based on current iterations. This is a default learning rate decay in Adafactor.


Float. If set, weight decay is applied.


Float. If set, the gradient of each weight is individually clipped so that its norm is no higher than this value.


Float. If set, the gradient of each weight is clipped to be no higher than this value.


Float. If set, the gradient of all weights is clipped so that their global norm is no higher than this value.


Boolean, defaults to FALSE. If TRUE, exponential moving average (EMA) is applied. EMA consists of computing an exponential moving average of the weights of the model (as the weight values change after each training batch), and periodically overwriting the weights with their moving average.


Float, defaults to 0.99. Only used if use_ema = TRUE. This is the momentum to use when computing the EMA of the model's weights: new_average = ema_momentum * old_average + (1 - ema_momentum) * current_variable_value.


Int or NULL, defaults to NULL. Only used if use_ema=TRUE. Every ema_overwrite_frequency steps of iterations, we overwrite the model variable by its moving average. If NULL, the optimizer does not overwrite model variables in the middle of training, and you need to explicitly overwrite the variables at the end of training by calling optimizer$finalize_variable_values() (which updates the model variables in-place). When using the built-in fit() training loop, this happens automatically after the last epoch, and you don't need to do anything.


String. The name to use for momentum accumulator weights created by the optimizer.


For forward/backward compatability.


Float or NULL. If a float, the scale factor will be multiplied the loss before computing gradients, and the inverse of the scale factor will be multiplied by the gradients before updating variables. Useful for preventing underflow during mixed precision training. Alternately, optimizer_loss_scale() will automatically set a loss scale factor.


Int or NULL. If an int, model & optimizer variables will not be updated at every step; instead they will be updated every gradient_accumulation_steps steps, using the average value of the gradients since the last update. This is known as "gradient accumulation". This can be useful when your batch size is very small, in order to reduce gradient noise at each update step.


an Optimizer instance