Arao AI
Back to search results

Qwen3-4B · Q5_K_M

Start a discussion
Apple M4 Pro64GBvLLM v0.45L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3-4B (Q5_K_M) on Apple M4 Pro with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

115.86 tok/s

Decode

Decode speed

428.93 tok/s

Prefill

Prefill speed

1.61 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

5 days ago

Last verified

Performance

  1. vLLM · Q5_K_M (current)Decode 115.86 · Prefill 428.93 ·
  2. Ollama · IQ2_XS Decode 164.02 · Prefill 802.68 ·
  3. Ollama · Q8_0 Decode 86.82 · Prefill 468.6 ·

Core figures

Decode (avg)
115.86 tok/s
Prefill (avg)
428.93 tok/s
TTFT (avg)
1.61 s
VRAM (avg)
9.91 GB
MTP acceptance rate
—
TTFB
0.93 GB
Power draw
66.3 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3-4B
Quantization
Q5_K_M
Framework
vLLM
Version
v0.45
Context length
4096 tokens
Batch size
16
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 M4 Pro
Nominal VRAM
64 GB
Measured VRAM (avg)
9.91 GB
OS
macOS 15.0
Driver
Apple M4 Pro
CUDA
N/A
Power draw
66.3 W

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-13

    v0.45 · macOS 15.0 · CUDA N/A · 4096 ctx

    115.86 tok/s

    Decode

    428.93 tok/s

    Prefill

    1.61 s

    TTFT

    9.91 GB

    VRAM

    —

    MTP

    66.3 W

    Power draw

    L2 batch seed 2026-10-13: vLLM on hw_id=25 running model_id=54 at Q5_K_M