perf: add GB200 DeepSeek V4 FP4 vLLM MTP AgentX - #2567

Open

cquil11 wants to merge 13 commits into

mainfrom

agent/gb200-dsv4-vllm-mtp-agentx

Commits on Aug 11, 2026

Add GB200 DeepSeek V4 vLLM MTP AgentX sweep
cquil11
committed
f10fcb6
Document GB200 DeepSeek V4 AgentX submission
cquil11
committed
f284554

Commits on Aug 12, 2026

Align GB200 DeepSeek V4 with vLLM recipe
cquil11
committed
6c2df11
Merge remote-tracking branch 'origin/main' into agent/gb200-dsv4-vllm-mtp-agentx
cquil11
committed
c49125e
Fix GB200 DeepSeek V4 MTP startup paths
cquil11
committed
4c98ed7
fix DeepSeek V4 MTP graph capture scope
cquil11
committed
0904351
fix(gb200): use current vLLM for DeepSeek V4 MTP
cquil11
committed
033d928
fix(gb200): reserve decode memory for MTP graphs
cquil11
committed
824b857
fix(gb200): report vLLM prompt cache usage
cquil11
committed
bac4588
revert(gb200): drop unsupported vLLM cache flag
cquil11
committed
1c225fe
fix(gb200): allow long DeepSeek prefill responses
cquil11
committed
50c1700
perf(gb200): broaden DeepSeek V4 AgentX search
cquil11
committed
b1f2928
Merge main and finalize GB200 DeepSeek V4 full sweep
cquil11
committed
38f342b