Arao AI
Back to search results

Qwen2.5-0.5B-Instruct · Q4_K_M

Start a discussion
Apple M1vLLM v0.48L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen2.5-0.5B-Instruct (Q4_K_M) on Apple M1 with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

143.97 tok/s

Decode

Decode speed

610.72 tok/s

Prefill

Prefill speed

2.02 s

TTFT

Time to first token

2.96 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

2 days ago

Last verified

Performance

  1. llama.cpp · IQ2_M Decode 189.83 · Prefill 1001.05 ·
  2. llama.cpp · Q2_K Decode 189.43 · Prefill 860.24 ·
  3. vLLM · Q3_K_S Decode 170.33 · Prefill 911.66 ·
  4. vLLM · Q4_K_M (current)Decode 143.97 · Prefill 610.72 ·

Core figures

Decode (avg)
143.97 tok/s
Prefill (avg)
610.72 tok/s
TTFT (avg)
2.02 s
VRAM (avg)
2.96 GB
MTP acceptance rate
—
TTFB
1.15 GB
Power draw
64 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen2.5-0.5B-Instruct
Quantization
Q4_K_M
Framework
vLLM
Version
v0.48
Context length
16384 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 M1
Measured VRAM (avg)
2.96 GB
OS
macOS 15.0
Driver
Apple M1
CUDA
N/A
Power draw
64 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

    v0.48 · macOS 15.0 · CUDA N/A · 16384 ctx

    143.97 tok/s

    Decode

    610.72 tok/s

    Prefill

    2.02 s

    TTFT

    2.96 GB

    VRAM

    —

    MTP

    64 W

    Power draw

    L2 batch seed 2026-10-16: vLLM on hw_id=99 running model_id=59 at Q4_K_M