@claude use `vllm/vllm-openai:v0.15.1-cu130` for TP=2 & TP=4 from conc=4 to 64 on minimax model `MiniMaxAI/MiniMax-M2.5`
Activity
Klaud-Cold commentedon Feb 19, 2026
Claude finished @functionstackx's task in 69m 32s —— View job •
claude/issue-756-20260219-0235• Create PR ➔Task: Add B200 MiniMax M2.5 FP8 vLLM Benchmarks
minimaxm2.5_fp8_b200.shminimaxm2.5-fp8-b200-vllmto nvidia-master.yamlChanges
New files:
benchmarks/minimaxm2.5_fp8_b200.sh— vLLM benchmark script for MiniMax-M2.5 FP8 on B200 (adapted from MI355X script, removing ROCm-specific config)Modified files:
.github/configs/nvidia-master.yaml— Addedminimaxm2.5-fp8-b200-vllmconfig:vllm/vllm-openai:v0.15.1-cu130MiniMaxAI/MiniMax-M2.5perf-changelog.yaml— Added changelog entryE2E Test Status
Smoke test running (TP=2,4 conc=4 1k1k): Run #22166363309
b200-nb_1(model download + server startup for 456B MoE model)|
Add MiniMax-M2.5 FP8 vLLM benchmark for B200
Add MiniMax-M2.5 FP8 vLLM benchmark for B200
Add MiniMax-M2.5 FP8 vLLM benchmark for B200 (#757)