vLLM Shares P/D Prototype and MTP3 Benchmark Results

vLLM shared pre-launch benchmarks: its P/D (prefill/decode disaggregation) prototype on 12 GB300 GPUs achieved a 91.24% prefix hit rate, while MTP3 validation on B300 DEP8 raised AIME25 from 0.9500 to 0.9667 with zero errors.

2026-10-15 ~ 2026-10-15 · 2 related posts

• vLLM MTP3 validation: AIME25 up to 0.9667 with prefill latency cuts — vllm_project · 2026-10-15
• vLLM P/D prototype hits 91.24% prefix cache hit rate on 12× GB300 — vllm_project · 2026-10-15