Qwen3.6-27B · vLLM · NVFP4 · conc 4
status: done
Configuration
| Model | Qwen/Qwen3.6-27B |
|---|---|
| Company | Alibaba |
| Family | Qwen |
| Parameters | 27B (dense) |
| Engine | vLLM |
| Quant / precision | NVFP4 |
| Why this quant | conc-4 base (non-spec) point of the Qwen3.6-27B NVFP4 sweep — matched no-spec baseline for the MTP conc-4 row (EXPERIMENTS.md |
| Download | unsloth/Qwen3.6-27B-NVFP4 |
| Context window | 65536 |
| Input modalities | text, image, video (served text-only here) |
Measured results
| Prefill tok/s | 56.89 |
|---|---|
| Decode tok/s | 35.54 |
| Peak memory (GB) | 106.39 (system MemAvailable delta (10s sampling) — vLLM static KV reservation (util 0.85)) |
| Completed | 2026-07-01 22:16 +0800 |
Full run command
# conc-4 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 4 1000 600 256 \
--trust-remote-code --dtype bfloat16
Notes
conc-4 base (no-spec) point of the Qwen3.6-27B NVFP4 sweep — matched baseline for the MTP conc-4 row (EXPERIMENTS.md #4/#14).
- Result (conc 4): prefill 56.89 / decode 35.54 tok/s aggregate; 0 errors; peak mem 106.4 GB.
- Still near-linear (9.33 → 18.11 → 35.54 at c1/2/4 ≈ ×1.94/×1.96) — the dense 27B stays bandwidth-bound
through conc-4, so the batch buys almost proportional aggregate decode. MTP-vs-base delta on the
-mtp-c4page. - Sweep siblings (base):
-c1·-c2·-c8·-c16·c32(main).