hstu

Classes

HSTUModel(n_blocks, n_heads, n_factors, ...)	HSTU model: transformer-based sequential model with unidirectional pointwise aggregated attention mechanism, combined with "Shifted Sequence" training objective.
HSTUModelConfig(*, cls, verbose, ...)	HSTU model config.
RelativeAttentionBias(session_max_len, ...)	Computes relative time and positional attention biases for STU.
STULayer(n_factors, n_heads, ...)	HSTU author's encoder block architecture rewritten from jagged tensor to dense.
STULayers(n_blocks, n_factors, n_heads, ...)	STULayers transformer blocks.