Like dict.update, adds/replaces values from other.

tf.zeros().

Accumulate histogram of binned statistic by label. Args: statistics: int32 np.array of shape [K, 1] of binned statistic labels: int32

Adds tensors in nmap_x with respective tensors in nmap_y. Args: nmap_x: A `.NestedMap` of tensors. nmap_y: A `.NestedMap` of tensors. Re

Augments state0 with additional state.

Returns the FileWriter object to use for summaries.

Imports the python module of the given name.

Returns whether an object is an instance of a collections.namedtuple. Examples:: _IsNamedTuple((42, 'hi')) ==> False Foo = collections.nam

Applies layer normalization on the last dimension of 'x'. Args: theta: a NestedMap of layer params. gate_name: the name of the gate,

Attempt to consume the given symbol. The symbol must be a member of the symbol set: []()=, Args: symbol: The symbol that we expect at

Performs a softmax as if padding were applied after exponentiation. The default implementation uses numerical techniques to approximate this

Helper to construct a parallel layer which merge branches.

Helper for '->rb' Graph edge.

Smoothed L1 norm.

Returns dimensions commonly used in StreamStep methods. Args: inputs: The query_vec parameter of StreamStep. Returns: A NestedMa

Call async thread with default Error callback to break main.

Create TensorArray from v, of size max_seq_length.

Returns partitioned tensors.

Add existing layer or layers as sublayer.

Adds a metric to the eval metrics. Args: name: A python string. The name of the metric. value: A scalar Tensor. weight: A scala

Add framework tensors like step_seed and global_step to inputs. This is only necessary when using `func`, as wrapping is handled automaticall

Assigns anchors to bboxes using a similarity function (SSD-based). Each anchor box is assigned to the top matching ground truth box. Ground t

Compute the input layer.

Computes moments and updates state. Args: theta: A `.NestedMap` object containing weights' values of this layer and its children la

Same as the ComputeContextVectorWithSource api above, except values ... in source_vecs, source_contexts and source_padding are ordered differentl

Corner regularization loss. This function computes the corner loss, an alternative regression loss for box residuals. This was used in the Fr

Returns the current cluster specification. E.g.:: with Cluster(cluster_params) as foo: model = p.Instantiate() model.FProp() # FP

TransformerDecoder SelfAttention. Note that attention bias (see _DecNotvisible) ensures that current position (~row) is less that memory posi

Create CPU dequeue ops. Args: concat: Whether to concat the passthrough batches for each host into one batch. Returns: N

Set up device-loop-level params.

Disables the accumulator (must be balanced with Enable).

Enables the accumulator (must balance a Disable).

Extend cached source_vecs and source_contexts by one more timestep. Args: theta: A `.NestedMap` object containing weights' values of this l

Transformer Layer, extend one step in decoding. This function is expected to be called during fast decoding of Transformer models. Args:

Extracts Tensors from a tf.Example record using self.extractors. Args: record: A tf.Example input to pass to tf.io.parse_single_example.

Forward pass. Args: theta: A NestedMap object containing weights' values of this layer and its children layers. inputs: A Nes

Round-robin every children cells in cell_tpl among worker devices. Args: theta: A NestedMap object containing weights' values of this layer

Forward function. The default implementation here assumes the cell forward function is composed of two functions:: _Gates(_Mix(theta

Applies SymbolInsertionLayer. We take in a `x`, which represents the groundtruth sequence (i.e., English sequence). We return a sampled rolli

GIN-based Featurizer for Model V2.

Return the list of shape prefixes for CreateVariable().

Returns the total batch size (for stats), int or dynamic int tensor.

Returns the batch size of the input batch: int or dynamic int tensor.

Returns a list of all the keys in nested/array-style form if needed.

Linear layer for predicting residuals and classification logits.