[Spec] Point multi-layer eagle's last shared-read runner at the draft runner (#35057)
This commit is contained in:
@@ -955,6 +955,11 @@ class MultiLayerEagleWorkerV2(BaseSpecWorker):
|
|||||||
|
|
||||||
self.plan_stream, self.plan_stream_ctx = get_plan_stream(self.device)
|
self.plan_stream, self.plan_stream_ctx = get_plan_stream(self.device)
|
||||||
|
|
||||||
|
@property
|
||||||
|
def last_shared_read_runner(self):
|
||||||
|
# Multi-layer eagle has no draft forward, only draft extend.
|
||||||
|
return self._draft_worker.draft_runner
|
||||||
|
|
||||||
@property
|
@property
|
||||||
def spec_v2_attn_backends(self) -> tuple:
|
def spec_v2_attn_backends(self) -> tuple:
|
||||||
return (
|
return (
|
||||||
|
|||||||
Reference in New Issue
Block a user