Fix data parallel + tensor parallel (#4499)

This commit is contained in:
Lianmin Zheng
2025-03-17 05:13:16 -07:00
committed by GitHub
parent f2ab37e500
commit 5493c3343e
6 changed files with 53 additions and 16 deletions

View File

@@ -1786,7 +1786,7 @@ def run_scheduler_process(
prefix = f" DP{dp_rank} TP{tp_rank}"
# Config the process
# kill_itself_when_parent_died() # This is disabled because it does not work for `--dp 2`
kill_itself_when_parent_died()
setproctitle.setproctitle(f"sglang::scheduler{prefix.replace(' ', '_')}")
faulthandler.enable()
parent_process = psutil.Process().parent()