AI terms starting with S
Every term in this glossary whose name begins with S, each with a one-paragraph definition you can read without opening the page.
Terms
AI terms starting with S
sampling
How a model picks the next token: draw from the softmax over the vocabulary, usually after a temperature rescale and a top-p or top-k truncation.
session
A fiction your code maintains: the history, budgets and tool state you reassemble and resend on every request.
speculative decoding
An inference technique where a small draft model proposes several tokens and the large target model verifies them in one forward pass, cutting latency without changing the target output distribution.
stop sequence
A string that halts decoding when the model would emit it. The completion excludes the match, and the response's stop reason names which sequence fired.
streaming
A response mode where the model returns token deltas over one HTTP connection as it generates, so your client can render, inspect or cancel before the answer is complete.
structured outputs
A provider feature that constrains the model's next-token sampling to tokens that keep a partial JSON parse valid against your schema, so required fields, types and enum values are guaranteed.
subagent
A separate agent a parent spawns for a subtask, with its own context and tools; only the final message returns to the parent.