pylibwholegraph.torch.embedding.WholeMemoryOptimizer

class pylibwholegraph.torch.embedding.WholeMemoryOptimizer(global_comm: WholeMemoryCommunicator)

Sparse Optimizer for WholeMemoryEmbedding. Many WholeMemoryEmbedding can share same WholeMemoryOptimizer You should not create WholeMemoryOptimizer object directly, but use create_wholememory_optimizer() instead.

Methods

add_embedding(wm_embedding)	Add WholeMemory Embedding to this optimizer NOTE: you don't need to call this method, it is automatic called when WholeMemory Optimizer is created. :param wm_embedding: WholeMemory Embedding that use this optimizer :return: None.
step(lr)	Apply gradients to all WholeMemory Embedding that use this optimizer.

__init__(global_comm: WholeMemoryCommunicator)

Methods

__init__(global_comm)
add_embedding(wm_embedding)	Add WholeMemory Embedding to this optimizer NOTE: you don't need to call this method, it is automatic called when WholeMemory Optimizer is created. :param wm_embedding: WholeMemory Embedding that use this optimizer :return: None.
step(lr)	Apply gradients to all WholeMemory Embedding that use this optimizer.