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

[server_args] Make MoRI per-rank dispatch-token requirement overridable (#29096)

Browse Source
This commit is contained in:
Chengze Fan
2026-06-26 22:25:36 -07:00
committed by GitHub
parent 495f13fa12
commit 19abebcc6a
1 changed files with 10 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/server_args.py
+10 -2
View File
@@ -5548,8 +5548,16 @@ class ServerArgs:
# Skip validation if disaggregation mode is decode.
if self.chunked_prefill_size > 0 and self.disaggregation_mode != "decode":
assert (
self.chunked_prefill_size
) <= envs.SGLANG_MORI_NUM_MAX_DISPATCH_TOKENS_PER_RANK.get(), "SGLANG_MORI_NUM_MAX_DISPATCH_TOKENS_PER_RANK (default 4096) must be larger or equal to chunked_prefill_size"
self._required_mori_dispatch_tokens_per_rank()
) <= envs.SGLANG_MORI_NUM_MAX_DISPATCH_TOKENS_PER_RANK.get(), (
"SGLANG_MORI_NUM_MAX_DISPATCH_TOKENS_PER_RANK (default 4096) "
"must be >= the per-rank MoRI dispatch tokens "
"(chunked_prefill_size by default)"
)
def _required_mori_dispatch_tokens_per_rank(self) -> int:
"""Max tokens a single rank dispatches through MoRI in one forward."""
return self.chunked_prefill_size
def _handle_eplb_and_dispatch(self):
if self.enable_eplb and (self.expert_distribution_recorder_mode is None):