[Ascend] Support enable-mixed-chunk in non-MLA scenarios (#12491)

Co-authored-by: ronnie_zheng <zl19940307@163.com>
2025-11-26 08:00:21 -08:00
parent 6c190cbda0
commit 262c3c1fde
3 changed files with 262 additions and 38 deletions
--- a/python/sglang/srt/managers/schedule_policy.py
+++ b/python/sglang/srt/managers/schedule_policy.py
@@ -553,7 +553,7 @@ class PrefillAdder:
                min(req.sampling_params.max_new_tokens, CLIP_MAX_NEW_TOKENS),
            )
        else:
-            if self.rem_chunk_tokens == 0:
+            if self.rem_chunk_tokens <= 0:
                return AddReqResult.OTHER

            # Chunked prefill