Arao AI
Back to search results

Qwen3-4B-Thinking-2507 · Q3_K_S

Start a discussion
NVIDIA Jetson AGX Orin 64GB64GBvLLM v0.52L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3-4B-Thinking-2507 (Q3_K_S) on NVIDIA Jetson AGX Orin 64GB with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

89.8 tok/s

Decode

Decode speed

485.57 tok/s

Prefill

Prefill speed

1.3 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

2 days ago

Last verified

Performance

  1. vLLM · Q3_K_S (current)Decode 89.8 · Prefill 485.57 ·
  2. vLLM · Q5_K_S Decode 57.15 · Prefill 278.01 ·
  3. Ollama · Q2_K Decode 77.04 · Prefill 283.31 ·
  4. llama.cpp · IQ2_XS Decode 92.94 · Prefill 357.53 ·
  5. Strata · Q5_K_S Decode 83.78 · Prefill 413.16 ·

Core figures

Decode (avg)
89.8 tok/s
Prefill (avg)
485.57 tok/s
TTFT (avg)
1.3 s
VRAM (avg)
7.56 GB
MTP acceptance rate
—
TTFB
0.72 GB
Power draw
49.2 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3-4B-Thinking-2507
Quantization
Q3_K_S
Framework
vLLM
Version
v0.52
Context length
32768 tokens
Batch size
8
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 Jetson AGX Orin 64GB
Nominal VRAM
64 GB
Measured VRAM (avg)
7.56 GB
OS
JetPack 6.0
Driver
NVIDIA L4T 36.3
CUDA
12.6
Power draw
49.2 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

    v0.52 · JetPack 6.0 · CUDA 12.6 · 32768 ctx

    89.8 tok/s

    Decode

    485.57 tok/s

    Prefill

    1.3 s

    TTFT

    7.56 GB

    VRAM

    —

    MTP

    49.2 W

    Power draw

    L2 batch seed 2026-10-16: vLLM on hw_id=76 running model_id=57 at Q3_K_S