[Spec v2] Remove allocate_lens and enable over-allocation (#13478)
This commit is contained in:
@@ -1760,7 +1760,7 @@ class ScheduleBatch(ScheduleBatchDisaggregationDecodeMixin):
|
||||
|
||||
def merge_batch(self, other: "ScheduleBatch"):
|
||||
# NOTE: in v2 eagle mode, we do not need wait verify here because
|
||||
# 1) current batch is always prefill, whose seq_lens and allocate_lens are not a future
|
||||
# 1) current batch is always prefill, whose seq_lens is not a future
|
||||
# 2) other batch is always decode, which is finished in previous step
|
||||
|
||||
# Penalizer orchestrator must be merged before Batch.reqs is merged. This is because
|
||||
|
||||
Reference in New Issue
Block a user