[Deepseek V3.2] Only skip Indexer logits computation when is_extend_without_speculative (#12816)

Signed-off-by: Hao Lu <14827759+hlu1@users.noreply.github.com>
This commit is contained in:
hlu1
2025-11-07 01:34:04 -08:00
committed by GitHub
parent bc25ea6762
commit bef37d6de8
3 changed files with 20 additions and 18 deletions

View File

@@ -141,6 +141,13 @@ class ForwardMode(IntEnum):
def is_split_prefill(self):
return self == ForwardMode.SPLIT_PREFILL
def is_extend_without_speculative(self):
return (
self.is_extend()
and not self.is_target_verify()
and not self.is_draft_extend()
)
@total_ordering
class CaptureHiddenMode(IntEnum):