Add overlap scheduling for embeddings code path (#14032)

Co-authored-by: Satyam Kumar <satyamk@linkedin.com>
This commit is contained in:
satyamk7054
2025-12-24 18:24:18 -08:00
committed by GitHub
co-authored by Satyam Kumar
parent ecd2d09a6c
commit 38dd4fbb66
6 changed files with 63 additions and 19 deletions
@@ -224,6 +224,9 @@ class SchedulerOutputProcessorMixin:
)
else: # embedding or reward model
if result.copy_done is not None:
result.copy_done.synchronize()
is_sparse = envs.SGLANG_EMBEDDINGS_SPARSE_HEAD.is_set()
embeddings = result.embeddings