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
EagleDraftAdapter
EagleDraftAdapter
class max.pipelines.speculative.eagle.EagleDraftAdapter(draft, hidden_dim)
Bases: EagleProposer
Adapts a registered EAGLE’s typed calls to the existing driver.
prefill()
prefill(batch, tokens, target_hidden)
Runs the initial draft call over the corrected target window.
-
Parameters:
-
- batch (SequentialBatch)
- tokens (TensorValue)
- target_hidden (TensorValue)
-
Return type:
step()
step(batch, draft_input, index)
Runs a later draft call using the previous hidden carry.
-
Parameters:
-
- batch (SequentialBatch)
- draft_input (DraftStepInput)
- index (int)
-
Return type: