minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

Fix LongCat n-gram embedding in PD-disaggregated scheduler loops (#31134)

Browse Source 
Co-authored-by: whn09 <whn09@users.noreply.github.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
王鹤男
2026-07-16 15:42:51 +08:00
committed by GitHub
co-authored by whn09 Claude Opus 4.8
parent a614821341
commit 5cbea10e2f
2 changed files with 12 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/disaggregation/decode.py
+6
View File
@@ -2001,6 +2001,9 @@ class SchedulerDisaggregationDecodeMixin:
) )
self.running_batch = plan.running_batch self.running_batch = plan.running_batch
batch = plan.batch_to_run batch = plan.batch_to_run
batch = self.ngram_embedding_manager.prepare_for_forward(
batch, chunked_req=self.chunked_req
)
self.cur_batch_for_debug = batch self.cur_batch_for_debug = batch
# Launch the current batch # Launch the current batch
@@ -2039,6 +2042,9 @@ class SchedulerDisaggregationDecodeMixin:
) )
self.running_batch = plan.running_batch self.running_batch = plan.running_batch
batch = plan.batch_to_run batch = plan.batch_to_run
batch = self.ngram_embedding_manager.prepare_for_forward(
batch, chunked_req=self.chunked_req
)
self.cur_batch_for_debug = batch self.cur_batch_for_debug = batch
# overlap + spec + grammar is unsupported (would desync DP ranks). # overlap + spec + grammar is unsupported (would desync DP ranks).
disable_overlap_for_batch = self.is_disable_overlap_for_batch( disable_overlap_for_batch = self.is_disable_overlap_for_batch(
python/sglang/srt/disaggregation/prefill.py
+6
View File
@@ -521,6 +521,9 @@ class SchedulerDisaggregationPrefillMixin:
) )
self.running_batch = plan.running_batch self.running_batch = plan.running_batch
batch = plan.batch_to_run batch = plan.batch_to_run
batch = self.ngram_embedding_manager.prepare_for_forward(
batch, chunked_req=self.chunked_req
)
self.cur_batch_for_debug = batch self.cur_batch_for_debug = batch
# Launch the current batch # Launch the current batch
@@ -559,6 +562,9 @@ class SchedulerDisaggregationPrefillMixin:
) )
self.running_batch = plan.running_batch self.running_batch = plan.running_batch
batch = plan.batch_to_run batch = plan.batch_to_run
batch = self.ngram_embedding_manager.prepare_for_forward(
batch, chunked_req=self.chunked_req
)
self.cur_batch_for_debug = batch self.cur_batch_for_debug = batch
# Launch the current batch # Launch the current batch