Arao AI
Back to search results

Qwen3-4B-Thinking-2507 · Q5_K_M

Start a discussion
NVIDIA DGX Spark128GBvLLM v0.55L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3-4B-Thinking-2507 (Q5_K_M) on NVIDIA DGX Spark with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

151.71 tok/s

Decode

Decode speed

832.95 tok/s

Prefill

Prefill speed

2.16 s

TTFT

Time to first token

9.91 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

2 days ago

Last verified

Performance

  1. vLLM · IQ1_S Decode 291.5 · Prefill 1360.04 ·
  2. vLLM · Q5_K_M (current)Decode 151.71 · Prefill 832.95 ·
  3. Strata · Q5_K_M Decode 220.61 · Prefill 780.86 ·
  4. Ollama · Q3_K_L Decode 191.76 · Prefill 876.65 ·
  5. Ollama · Q5_K_S Decode 148.12 · Prefill 634.82 ·

Core figures

Decode (avg)
151.71 tok/s
Prefill (avg)
832.95 tok/s
TTFT (avg)
2.16 s
VRAM (avg)
9.91 GB
MTP acceptance rate
—
TTFB
1.21 GB
Power draw
165.6 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3-4B-Thinking-2507
Quantization
Q5_K_M
Framework
vLLM
Version
v0.55
Context length
16384 tokens
Batch size
16
GPU layers
999
Flash Attention
On

Hardware

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

GPU
NVIDIA DGX Spark
Nominal VRAM
128 GB
Measured VRAM (avg)
9.91 GB
OS
Ubuntu 22.04
Driver
570.86
CUDA
12.4
Power draw
165.6 W

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-16

    v0.55 · Ubuntu 22.04 · CUDA 12.4 · 16384 ctx

    151.71 tok/s

    Decode

    832.95 tok/s

    Prefill

    2.16 s

    TTFT

    9.91 GB

    VRAM

    —

    MTP

    165.6 W

    Power draw

    L2 batch seed 2026-10-16: vLLM on hw_id=54 running model_id=57 at Q5_K_M