Tune B200 DSV4 SGLang AgentX prefill scheduling / 调整 B200 DSV4 SGLang AgentX 预填充调度 - #2718

Merged

cquil11 merged 14 commits into

mainfrom

dsv4-fp4-b200-sglang-agentic-prefill-scheduling

Aug 28, 2026

Commits on Aug 28, 2026

Tune B200 DSV4 AgentX prefill scheduling
nvpohanh
committed
10fc322
Update perf-changelog PR link
nvpohanh
committed
0868b3d
Fix perf-changelog PR backfill
nvpohanh
committed
aa53a57
perf(dsv4): refresh B200 AgentX high-concurrency curve
nvpohanh
committed
cec0d51
refactor(dsv4): simplify B200 AgentX flag dispatch
nvpohanh
committed
7665319
fix(dsv4): restore B200 indexer memory headroom
nvpohanh
committed
36d0781
fix(dsv4): preserve c96 FP4 indexer headroom
nvpohanh
committed
e75da5a
fix(dsv4): fit TP HiCache on NScale
nvpohanh
committed
e7a6ee4
fix(dsv4): restore c160 indexer headroom
nvpohanh
committed
014e993
refactor(dsv4): unify DP workspace limits
nvpohanh
committed
04009c0
refactor(dsv4): unify DEP8 request routing
nvpohanh
committed
4b65ec3
chore(dsv4): update B200 SGLang nightly image
nvpohanh
committed
6bff0ff
chore(dsv4): roll back B200 SGLang nightly
nvpohanh
committed
c705ac5
fix(dsv4): reserve indexer workspace for B200 DP
nvpohanh
committed
30f2fb4