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

fix: Auto-correct page_size for Mamba no_buffer radix cache mode (#20538)

Browse Source
This commit is contained in:
shuwenn
2026-04-08 00:19:31 +08:00
committed by GitHub
parent 727a182067
commit ec5742f4ab
1 changed files with 7 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/server_args.py
+7
View File
@@ -2204,6 +2204,13 @@ class ServerArgs:
== 0 == 0
), f"For SSM models with extra buffer, either FLA_CHUNK_SIZE or page_size must be divisible by the other, got {FLA_CHUNK_SIZE=}, {self.page_size=}" ), f"For SSM models with extra buffer, either FLA_CHUNK_SIZE or page_size must be divisible by the other, got {FLA_CHUNK_SIZE=}, {self.page_size=}"
elif not self.disable_radix_cache: # no_buffer elif not self.disable_radix_cache: # no_buffer
if self.page_size is not None and self.page_size != 1:
logger.warning(
f"{model_arch} with radix cache requires page_size=1 in the current "
f"Mamba scheduling mode (no_buffer), but got {self.page_size}. "
"Automatically setting page_size=1."
)
self.page_size = 1
if self.speculative_algorithm is None: if self.speculative_algorithm is None:
logger.warning( logger.warning(
"Disabling overlap schedule since mamba no_buffer is not compatible with " "Disabling overlap schedule since mamba no_buffer is not compatible with "