chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 11 Packages Projects Releases Wiki Activity

[Feature] Enable return routed experts (#12162)

Browse Source 
Co-authored-by: yizhang2077 <1109276519@qq.com>
Co-authored-by: Liangsheng Yin <lsyincs@gmail.com>
This commit is contained in:
Junrong Lin
2025-12-21 15:16:43 +08:00
committed by GitHub
co-authored by yizhang2077 Liangsheng Yin
parent 8fe3e37468
commit bed301a5ac
27 changed files with 646 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/managers/tp_worker.py
+1 -9
View File
@@ -317,15 +317,7 @@ class TpModelWorker(BaseTpWorker):
# Profile number of tokens
self.max_total_num_tokens = self.model_runner.max_total_num_tokens
self.max_prefill_tokens = server_args.max_prefill_tokens
self.max_running_requests = min(
(
self.max_total_num_tokens // 2
if server_args.max_running_requests is None
else server_args.max_running_requests
// (server_args.dp_size if server_args.enable_dp_attention else 1)
),
self.model_runner.req_to_token_pool.size,
)
self.max_running_requests = self.model_runner.max_running_requests
assert self.max_running_requests > 0, "max_running_request is zero"
self.max_queued_requests = server_args.max_queued_requests
assert (