Standalone speculative decoding (#10090)

2025-09-07 20:55:09 -07:00
parent 400d3b97ae
commit 8cda5a622c
11 changed files with 285 additions and 9 deletions
--- a/python/sglang/srt/managers/schedule_batch.py
+++ b/python/sglang/srt/managers/schedule_batch.py
@@ -1539,7 +1539,7 @@ class ScheduleBatch(ScheduleBatchDisaggregationDecodeMixin):
        self.forward_mode = ForwardMode.DECODE
        bs = len(self.reqs)

-        if self.spec_algorithm.is_eagle():
+        if self.spec_algorithm.is_eagle() or self.spec_algorithm.is_standalone():
            # if spec decoding is used, the decode batch is prepared inside
            # `forward_batch_speculative_generation` after running draft models.
            return