Arao AI
Back to search results

gemma-4-12B-it · Q4_K_S

Start a discussion
Apple M4 Pro64GBvLLM v0.30L2 Cross-framework verified

This page aggregates 1 real-world runs of gemma-4-12B-it (Q4_K_S) on Apple M4 Pro with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

90.85 tok/s

Decode

Decode speed

484.54 tok/s

Prefill

Prefill speed

0.68 s

TTFT

Time to first token

21.76 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

5 days ago

Last verified

Performance

  1. llama.cpp · IQ2_M Decode 103.7 · Prefill 452.45 ·
  2. llama.cpp · Q5_K_S Decode 68.46 · Prefill 358.45 ·
  3. vLLM · Q2_K Decode 110.03 · Prefill 407.67 ·
  4. vLLM · Q4_K_S (current)Decode 90.85 · Prefill 484.54 ·

Core figures

Decode (avg)
90.85 tok/s
Prefill (avg)
484.54 tok/s
TTFT (avg)
0.68 s
VRAM (avg)
21.76 GB
MTP acceptance rate
—
TTFB
0.73 GB
Power draw
52.7 W

Configuration

Member-level fields are taken from the most recent run

Model
gemma-4-12B-it
Quantization
Q4_K_S
Framework
vLLM
Version
v0.30
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
Apple M4 Pro
Nominal VRAM
64 GB
Measured VRAM (avg)
21.76 GB
OS
macOS 15.0
Driver
Apple M4 Pro
CUDA
N/A
Power draw
52.7 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

    v0.30 · macOS 15.0 · CUDA N/A · 32768 ctx

    90.85 tok/s

    Decode

    484.54 tok/s

    Prefill

    0.68 s

    TTFT

    21.76 GB

    VRAM

    —

    MTP

    52.7 W

    Power draw

    L2 batch seed 2026-10-13: vLLM on hw_id=25 running model_id=65 at Q4_K_S