For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /get-started.md).
Python class
DraftStepInput
DraftStepInput
class max.pipelines.speculative.driver.DraftStepInput(tokens, hidden)
Bases: object
Loop-carried state between draft steps.
Every draft in the tree threads the previous step’s token and its per-device hidden state.
-
Parameters:
-
- tokens (TensorValue)
- hidden (list[TensorValue])
hidden
hidden: list[TensorValue]
tokens
tokens: TensorValue