tf.raw_ops.XlaSparseDenseMatmulCustomCombinerOnTcGradWithCsrInput 

tf.raw_ops.XlaSparseDenseMatmulCustomCombinerOnTcGradWithCsrInput(
    row_pointers,
    sorted_sample_ids,
    sorted_token_ids,
    sorted_pos_ids,
    sorted_gains,
    weights,
    preserved_valencies,
    preserved_vectors,
    preserved_weights,
    activation_gradients,
    tables,
    hyperparameters,
    combiner_weights_learning_rate,
    max_valency,
    num_weights,
    combiner_table_vjp_computation,
    combiner_weights_vjp_computation,
    optimizer_custom_computation,
    table_name,
    num_sparsecores_per_device=-1,
    name=None
)

row_pointers A Tensor of type int32.
sorted_sample_ids A Tensor of type int32.
sorted_token_ids A Tensor of type int32.
sorted_pos_ids A Tensor of type int32.
sorted_gains A Tensor of type float32.
weights A Tensor of type float32.
preserved_valencies A Tensor of type int32.
preserved_vectors A Tensor of type float32.
preserved_weights A Tensor of type float32.
activation_gradients A Tensor of type float32.
tables A list of at least 1 Tensor objects with type float32.
hyperparameters A list of at least 1 Tensor objects with type float32.
combiner_weights_learning_rate A Tensor of type float32.
max_valency An int that is >= 0.
num_weights An int that is >= 0.
combiner_table_vjp_computation A function decorated with @Defun.
combiner_weights_vjp_computation A function decorated with @Defun.
optimizer_custom_computation A function decorated with @Defun.
table_name A string.
num_sparsecores_per_device An optional int. Defaults to -1.
name A name for the operation (optional).

A tuple of Tensor objects (updated_tables, updated_weights).
updated_tables A list with the same length as tables of Tensor objects with type float32.
updated_weights A Tensor of type float32.