Arao AI
Back to search results

Qwen3.5-4B · Q3_K_S

Start a discussion
NVIDIA H2097GBvLLM v0.51L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.5-4B (Q3_K_S) on NVIDIA H20 with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

84.6 tok/s

Decode

Decode speed

335.18 tok/s

Prefill

Prefill speed

0.71 s

TTFT

Time to first token

7.56 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

17 days ago

Last verified

Performance

  1. vLLM · Q3_K_S (current)Decode 84.6 · Prefill 335.18 ·
  2. vLLM · Q4_K_M Decode 68.71 · Prefill 315.97 ·
  3. Strata · IQ4_XS Decode 97.77 · Prefill 362.6 ·

Core figures

Decode (avg)
84.6 tok/s
Prefill (avg)
335.18 tok/s
TTFT (avg)
0.71 s
VRAM (avg)
7.56 GB
MTP acceptance rate
—
TTFB
0.68 GB
Power draw
326.1 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.5-4B
Quantization
Q3_K_S
Framework
vLLM
Version
v0.51
Context length
32768 tokens
Batch size
32
GPU layers
999
Flash Attention
Off

Hardware

Nominal and measured figures are shown side by side; whether it runs is the reader's call

GPU
NVIDIA H20
Nominal VRAM
97 GB
Measured VRAM (avg)
7.56 GB
OS
Windows 11 24H2
Driver
565.57
CUDA
12.6
Power draw
326.1 W

Sources & evidence

1 measured records in total, each traceable to its original source

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-19

    v0.51 · Windows 11 24H2 · CUDA 12.6 · 32768 ctx

    84.6 tok/s

    Decode

    335.18 tok/s

    Prefill

    0.71 s

    TTFT

    7.56 GB

    VRAM

    —

    MTP

    326.1 W

    Power draw

    L2 batch seed 2026-10-19: vLLM on hw_id=56 running model_id=48 at Q3_K_S