* An internal exception, as a child of std::runtime_error, to carry the error code. */

The planner work out a series of plans iteratively. In each iteration, the planner tries to split the hottest shard and place the shards into

Abbreviated as ``sok.DynamicVariable``. A variable that allocates memory dynamically. Parameters ---------- dimension: int

Input reader for pre-processed Criteo data. Raw Criteo data is assumed to be preprocessed in the following way: 1. Missing values are re

Input reader for pre-processed Criteo data. Raw Criteo data is assumed to be preprocessed in the following way: 1. Missing values are re

Sequential multi-layer perceptron (MLP) block.

Sequential multi-layer perceptron (MLP) block.

Sequential multi-layer perceptron (MLP) block.

Sequential multi-layer perceptron (MLP) block.

The planner work out a series of plans iteratively. In each iteration, the planner tries to split the hottest shard and place the shards into

Containing the state of a sharding process. The plan iteratively update the sharding state based on a given heuristic and obtain solution

Containing the state of a sharding process. The plan iteratively update the sharding state based on a given heuristic and obtain solution

* Helper class for switching device */

Dot interaction layer. See theory in the DLRM paper: https://arxiv.org/pdf/1906.00091.pdf, section 2.1.3. Sparse activations and dense activat

Dot interaction layer. See theory in the DLRM paper: https://arxiv.org/pdf/1906.00091.pdf, section 2.1.3. Sparse activations and dense activat

Dot interaction layer. See theory in the DLRM paper: https://arxiv.org/pdf/1906.00091.pdf, section 2.1.3. Sparse activations and dense activat

this unit test is same with test_embedding_collection.cpp this unit test is add embedding load dump during iteration will remove to a new unit test so

LR Scheduler combining Polynomial Decay with Warmup at the beginning. TF-based cond operations necessary for performance in graph mode.

LR Scheduler combining Polynomial Decay with Warmup at the beginning. TF-based cond operations necessary for performance in graph mode.