optax.lamb

Contents

optax.lamb#

optax.lamb(learning_rate: base.ScalarOrSchedule, b1: jax.typing.ArrayLike = 0.9, b2: jax.typing.ArrayLike = 0.999, eps: jax.typing.ArrayLike = 1e-06, eps_root: jax.typing.ArrayLike = 0.0, weight_decay: base.ScalarOrSchedule = 0.0, mask: MaskOrFn = None) → base.GradientTransformationExtraArgs[source]#

The LAMB optimizer.

LAMB is a general purpose layer-wise adaptive large batch optimizer designed to provide consistent training performance across a wide range of tasks, including those that use attention-based models (such as Transformers) and ResNet-50. The optimizer is able to work with small and large batch sizes. LAMB was inspired by the LARS learning algorithm.

Parameters:

  • learning_rate – A global scaling factor, either fixed or evolving along iterations with a scheduler, see optax.scale_by_learning_rate().

  • b1 – Exponential decay rate to track the first moment of past gradients.

  • b2 – Exponential decay rate to track the second moment of past gradients.

  • eps – A small constant applied to denominator outside of the square root (as in the Adam paper) to avoid dividing by zero when rescaling.

  • eps_root – A small constant applied to denominator inside the square root (as in RMSProp), to avoid dividing by zero when rescaling. This is needed for instance when computing (meta-)gradients through Adam.

  • weight_decay – Strength of the weight decay regularization.

  • mask – A tree with same structure as (or a prefix of) the params PyTree, or a Callable that returns such a pytree given the params/updates. The leaves should be booleans, True for leaves/subtrees you want to apply the transformation to, and False for those you want to skip.

Returns:

The corresponding optax.GradientTransformationExtraArgs.

Examples

>>> import optax
>>> import jax
>>> import jax.numpy as jnp
>>> def f(x): return jnp.sum(x ** 2)  # simple quadratic function
>>> solver = optax.lamb(learning_rate=0.003)
>>> params = jnp.array([1., 2., 3.])
>>> print('Objective function: ', f(params))
Objective function:  14.0
>>> opt_state = solver.init(params)
>>> for _ in range(5):
...  grad = jax.grad(f)(params)
...  updates, opt_state = solver.update(grad, opt_state, params)
...  params = optax.apply_updates(params, updates)
...  print('Objective function: {:.2E}'.format(f(params)))
Objective function: 1.39E+01
Objective function: 1.38E+01
Objective function: 1.38E+01
Objective function: 1.37E+01
Objective function: 1.36E+01

References

You et al, Large Batch Optimization for Deep Learning: Training BERT in 76 minutes, 2020

Contents