Arao AI
Back to search results

Qwen3-4B-Thinking-2507 · IQ3_S

Start a discussion
Apple M1 ProvLLM v0.43L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3-4B-Thinking-2507 (IQ3_S) on Apple M1 Pro with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

94.49 tok/s

Decode

Decode speed

403.53 tok/s

Prefill

Prefill speed

2.34 s

TTFT

Time to first token

7.64 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 · IQ3_S (current)Decode 94.49 · Prefill 403.53 ·
  2. Strata · IQ1_S Decode 163.46 · Prefill 649.89 ·
  3. Strata · Q2_K Decode 151.41 · Prefill 638.55 ·

Core figures

Decode (avg)
94.49 tok/s
Prefill (avg)
403.53 tok/s
TTFT (avg)
2.34 s
VRAM (avg)
7.64 GB
MTP acceptance rate
—
TTFB
0.76 GB
Power draw
61.9 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3-4B-Thinking-2507
Quantization
IQ3_S
Framework
vLLM
Version
v0.43
Context length
2048 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
Apple M1 Pro
Measured VRAM (avg)
7.64 GB
OS
macOS 15.0
Driver
Apple M1 Pro
CUDA
N/A
Power draw
61.9 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

    v0.43 · macOS 15.0 · CUDA N/A · 2048 ctx

    94.49 tok/s

    Decode

    403.53 tok/s

    Prefill

    2.34 s

    TTFT

    7.64 GB

    VRAM

    —

    MTP

    61.9 W

    Power draw

    L2 batch seed 2026-10-16: vLLM on hw_id=100 running model_id=57 at IQ3_S