Qwen3.6-27B · vLLM · NVFP4 · conc 2

status: done

Configuration

ModelQwen/Qwen3.6-27B
CompanyAlibaba
FamilyQwen
Parameters27B (dense)
EnginevLLM
Quant / precisionNVFP4
Why this quantconc-2 base (non-spec) point of the Qwen3.6-27B NVFP4 sweep — matched no-spec baseline for the MTP conc-2 row (EXPERIMENTS.md
Downloadunsloth/Qwen3.6-27B-NVFP4
Context window65536
Input modalitiestext, image, video (served text-only here)

Measured results

Prefill tok/s30.58
Decode tok/s18.11
Peak memory (GB)105.92 (system MemAvailable delta (10s sampling) — vLLM static KV reservation (util 0.85))
Completed2026-07-01 22:00 +0800

Full run command

# conc-2 base (no spec). Same recipe as the published conc-32 base — only --max-num-seqs differs.
scripts/bench-vllm-serving.sh unsloth/Qwen3.6-27B-NVFP4 65536 2 1000 600 256 \
  --trust-remote-code --dtype bfloat16

qwen3.6-27b Alibaba Qwen NVFP4 16-40B conc-2

Notes

conc-2 base (no-spec) point of the Qwen3.6-27B NVFP4 sweep — matched baseline for the MTP conc-2 row (EXPERIMENTS.md #4/#14).

  • Result (conc 2): prefill 30.58 / decode 18.11 tok/s aggregate; 0 errors; peak mem 105.9 GB.
  • Clean ~2× scaling from conc-1 (9.33 → 18.11) — the dense 27B is still bandwidth-bound at conc-2, so doubling the batch nearly doubles aggregate decode. MTP-vs-base delta is on the -mtp-c2 page.
  • Sweep siblings (base): -c1 · -c4 · -c8 · -c16 · c32 (main).