public final class LoadTPUEmbeddingMomentumParametersGradAccumDebug extends PrimitiveOp
An op that loads optimization parameters into HBM for embedding. Must be preceded by a ConfigureTPUEmbeddingHost op that sets up the correct embedding table configuration. For example, this op is used to install parameters that are loaded from a checkpoint before a training loop is executed.
Modifier and Type | Class and Description |
---|---|
static class |
LoadTPUEmbeddingMomentumParametersGradAccumDebug.Options
Optional attributes for
LoadTPUEmbeddingMomentumParametersGradAccumDebug |
operation
Modifier and Type | Method and Description |
---|---|
static LoadTPUEmbeddingMomentumParametersGradAccumDebug |
create(Scope scope,
Operand<Float> parameters,
Operand<Float> momenta,
Operand<Float> gradientAccumulators,
Long numShards,
Long shardId,
LoadTPUEmbeddingMomentumParametersGradAccumDebug.Options... options)
Factory method to create a class wrapping a new LoadTPUEmbeddingMomentumParametersGradAccumDebug operation.
|
static LoadTPUEmbeddingMomentumParametersGradAccumDebug.Options |
tableId(Long tableId) |
static LoadTPUEmbeddingMomentumParametersGradAccumDebug.Options |
tableName(String tableName) |
equals, hashCode, op, toString
public static LoadTPUEmbeddingMomentumParametersGradAccumDebug create(Scope scope, Operand<Float> parameters, Operand<Float> momenta, Operand<Float> gradientAccumulators, Long numShards, Long shardId, LoadTPUEmbeddingMomentumParametersGradAccumDebug.Options... options)
scope
- current scopeparameters
- Value of parameters used in the Momentum optimization algorithm.momenta
- Value of momenta used in the Momentum optimization algorithm.gradientAccumulators
- Value of gradient_accumulators used in the Momentum optimization algorithm.numShards
- shardId
- options
- carries optional attributes valuespublic static LoadTPUEmbeddingMomentumParametersGradAccumDebug.Options tableId(Long tableId)
tableId
- public static LoadTPUEmbeddingMomentumParametersGradAccumDebug.Options tableName(String tableName)
tableName
- Copyright © 2022. All rights reserved.