fix cpp JIT compilation issue of ngram speculative decoding (#10837)

2025-10-02 06:05:01 -07:00
parent 7d00479950
commit 948278f173
2 changed files with 4 additions and 1 deletions
--- a/python/sglang/srt/server_args.py
+++ b/python/sglang/srt/server_args.py
@@ -679,7 +679,7 @@ class ServerArgs:
                [1, 2, 4, 8, 12]
                + list(range(16, 257, 8))
                + list(range(272, 512, 16))
-                + list(range(512, self.cuda_graph_max_bs + 1))
+                + list(range(512, self.cuda_graph_max_bs + 1, 32))
            )
        else:
            # Spec decoding case: list(range(1, 9, 1)) + list(range(10, 33, 2)) + list(range(40, 64, 4)) + list(range(72, 257, 8))
--- a/python/sglang/srt/speculative/cpp_ngram/ngram.cpp
+++ b/python/sglang/srt/speculative/cpp_ngram/ngram.cpp
@@ -1,6 +1,9 @@
 #include "ngram.h"

+#include <algorithm>
+#include <cstring>
 #include <limits>
+#include <queue>
 #include <vector>

 namespace ngram {