[Bugfix] Stamp capture-time num_tokens_per_req in multi-layer EAGLE; close jit_kernel CI filter gaps (#31367)
Co-authored-by: Claude Fable 5 <noreply@anthropic.com> Co-authored-by: hnyls2002 <lsyincs@gmail.com>
This commit is contained in:
co-authored by
Claude Fable 5
hnyls2002
parent
3101c1258c
commit
7a973c03a0
@@ -273,6 +273,7 @@ class MultiLayerEagleDraftExtendCudaGraphRunner(DecodeCudaGraphRunner):
|
||||
num_correct_drafts=num_correct_drafts,
|
||||
num_accept_tokens=num_accept_tokens,
|
||||
)
|
||||
spec_info.num_tokens_per_req = self.num_tokens_per_req
|
||||
spec_info.positions = None
|
||||
|
||||
capture_mode = (
|
||||
|
||||
Reference in New Issue
Block a user