Validate vLLM Router on GB200: DEP4, DEP8, and 1P/2D - #2549

Open

cquil11 wants to merge 35 commits into

mainfrom

agent/vllm-router-gb200-smoke

Commits on Aug 10, 2026

feat: exercise native model routers through srt-slurm
cquil11
committed
f19290d
docs: register native router validation sweep
cquil11
committed
5c98e20
fix: verify vLLM Router through package metadata
cquil11
committed
53634d7
fix native router validation recipes
cquil11
committed
37243c4
fix: enable MNNVL for B200 vLLM router workers
cquil11
committed
d50adec
fix: use proven B200 vLLM communicator path
cquil11
committed
0bd3585
fix: use CUDA-aware UCX for SGL Router P/D
cquil11
committed
05b4761
chore: pin router startup-timeout support
cquil11
committed
d142e6e
test: move SGL Router P/D validation to H200
cquil11
committed
be0da91
fix: enable native routers on H200 launcher
cquil11
committed
62078c4
fix: pin native routers after SGL P/D warmup fix
cquil11
committed
31a4444
Validate native router jobs and pin PD fix
cquil11
committed
35e586b
Harden native router validation
cquil11
committed
32671d2
Merge remote-tracking branch 'origin/main' into agent/native-router-integration
cquil11
committed
6efbcfc
ci: recognize native router runtimes in sweep classifier
cquil11
committed
094b1ac
ci: harden native router sweep handling
cquil11
committed
a983e78
ci: pin latest validated srt-slurm router head
cquil11
committed
f58cd40
use semantic native router frontend names
cquil11
committed
20a1cd7
test official vLLM Router smoke points
cquil11
committed
95cc2cc
Merge remote-tracking branch 'origin/main' into agent/native-router-integration
cquil11
committed
406a6e4
test vLLM Router DEP4 on GB200
cquil11
committed
a5ebfb3
link GB200 router validation PR
cquil11
committed
da6c27a
test router-native DEP4 and DEP8 layouts
cquil11
committed
15c932f
Merge remote-tracking branch 'origin/main' into agent/vllm-router-gb200-smoke
cquil11
committed
56d4418
use session-affine AgentX router policy
cquil11
committed
8c66603
use 256k AgentX corpus for router validation
cquil11
committed
f323881
document 256k router validation context
cquil11
committed
59ed726
Merge remote-tracking branch 'origin/main' into agent/vllm-router-gb200-smoke
cquil11
committed
aa185a5
respect DeepSeek R1 context limit in router sweeps
cquil11
committed
93e5d56
pass explicit AgentX context limit to AIPerf
cquil11
committed
188bae0
reserve activation memory for router validation
cquil11
committed
395d57f

Commits on Aug 11, 2026