In the context of transformer based LLMs, a sequence refers to an ordered list of tokens.

A sequence contains the entire “window” of context that the model can see.

LLMs are constrained in the sequences it ingests based on the max sequence length.