Arao AI
Back to search results

DeepSeek-R1-Distill-Qwen-14B · Q3_K_S

Start a discussion
Apple M4 Pro64GBvLLM v0.52L2 Cross-framework verified

This page aggregates 1 real-world runs of DeepSeek-R1-Distill-Qwen-14B (Q3_K_S) on Apple M4 Pro with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

87.73 tok/s

Decode

Decode speed

414.18 tok/s

Prefill

Prefill speed

0.78 s

TTFT

Time to first token

21.21 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 · Q3_K_S (current)Decode 87.73 · Prefill 414.18 ·
  2. vLLM · Q4_K_M Decode 76.77 · Prefill 324.87 ·
  3. Strata · IQ3_S Decode 124.6 · Prefill 542.9 ·

Core figures

Decode (avg)
87.73 tok/s
Prefill (avg)
414.18 tok/s
TTFT (avg)
0.78 s
VRAM (avg)
21.21 GB
MTP acceptance rate
—
TTFB
0.7 GB
Power draw
101.3 W

Configuration

Member-level fields are taken from the most recent run

Model
DeepSeek-R1-Distill-Qwen-14B
Quantization
Q3_K_S
Framework
vLLM
Version
v0.52
Context length
8192 tokens
Batch size
32
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)
21.21 GB
OS
macOS 15.0
Driver
Apple M4 Pro
CUDA
N/A
Power draw
101.3 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

    v0.52 · macOS 15.0 · CUDA N/A · 8192 ctx

    87.73 tok/s

    Decode

    414.18 tok/s

    Prefill

    0.78 s

    TTFT

    21.21 GB

    VRAM

    —

    MTP

    101.3 W

    Power draw

    L2 batch seed 2026-10-13: vLLM on hw_id=25 running model_id=71 at Q3_K_S