[Bugfix][GDN] Reset speculative decode count for an empty draft schedule (#53077)
Signed-off-by: Kevin Luu <51931015+khluu@users.noreply.github.com>
This commit is contained in:
@@ -239,6 +239,7 @@ class GDNAttentionMetadataBuilder(AttentionMetadataBuilder[GDNAttentionMetadata]
|
||||
or num_decode_draft_tokens_cpu[spec_sequence_masks_cpu].sum().item()
|
||||
== 0
|
||||
):
|
||||
num_spec_decodes = 0
|
||||
spec_sequence_masks = None
|
||||
spec_sequence_masks_cpu = None
|
||||
else:
|
||||
|
||||
Reference in New Issue
Block a user