Arao AI
Back to search results

DeepSeek-R1-Distill-Qwen-14B · Q3_K_M

Start a discussion
Apple M5 Max128GBllama.cpp v0.22L2 Cross-framework verified

This page aggregates 1 real-world runs of DeepSeek-R1-Distill-Qwen-14B (Q3_K_M) on Apple M5 Max with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

66.61 tok/s

Decode

Decode speed

292.99 tok/s

Prefill

Prefill speed

1.37 s

TTFT

Time to first token

22.09 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

17 days ago

Last verified

Performance

  1. llama.cpp · Q3_K_M (current)Decode 66.61 · Prefill 292.99 ·
  2. Ollama · IQ4_XS Decode 63.06 · Prefill 323.43 ·
  3. Ollama · Q8_0 Decode 36.84 · Prefill 127.42 ·
  4. llama.cpp · IQ1_S Decode 202.22 · Prefill 672.97 ·
  5. vLLM · Q4_K_M Decode 134.81 · Prefill 469.71 ·

Core figures

Decode (avg)
66.61 tok/s
Prefill (avg)
292.99 tok/s
TTFT (avg)
1.37 s
VRAM (avg)
22.09 GB
MTP acceptance rate
—
TTFB
0.53 GB
Power draw
40.3 W

Configuration

Member-level fields are taken from the most recent run

Model
DeepSeek-R1-Distill-Qwen-14B
Quantization
Q3_K_M
Framework
llama.cpp
Version
v0.22
Context length
16384 tokens
Batch size
512
GPU layers
99
Flash Attention
On

Hardware

Nominal and measured figures are shown side by side; whether it runs is the reader's call

GPU
Apple M5 Max
Nominal VRAM
128 GB
Measured VRAM (avg)
22.09 GB
OS
macOS 26.1
Power draw
40.3 W

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-19

    v0.22 · macOS 26.1 · 16384 ctx

    66.61 tok/s

    Decode

    292.99 tok/s

    Prefill

    1.37 s

    TTFT

    22.09 GB

    VRAM

    —

    MTP

    40.3 W

    Power draw

    L2 batch seed 2026-10-19: llama.cpp on hw_id=28 running model_id=71 at Q3_K_M