Arao AI
Back to search results

Qwen2.5-Omni-3B · FP16

Start a discussion
Apple M1 Maxllama.cpp v0.52L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen2.5-Omni-3B (FP16) on Apple M1 Max with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

57.81 tok/s

Decode

Decode speed

312.83 tok/s

Prefill

Prefill speed

2.17 s

TTFT

Time to first token

13.76 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 · FP16 (current)Decode 57.81 · Prefill 312.83 ·
  2. llama.cpp · Q3_K_M Decode 128.76 · Prefill 591.62 ·
  3. Ollama · Q4_K_S Decode 116.7 · Prefill 504.6 ·

Core figures

Decode (avg)
57.81 tok/s
Prefill (avg)
312.83 tok/s
TTFT (avg)
2.17 s
VRAM (avg)
13.76 GB
MTP acceptance rate
—
TTFB
0.46 GB
Power draw
64.1 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen2.5-Omni-3B
Quantization
FP16
Framework
llama.cpp
Version
v0.52
Context length
4096 tokens
Batch size
512
GPU layers
20
Flash Attention
On

Hardware

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

GPU
Apple M1 Max
Measured VRAM (avg)
13.76 GB
OS
macOS 15.0
Driver
Apple M1 Max
CUDA
N/A
Power draw
64.1 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

    v0.52 · macOS 15.0 · CUDA N/A · 4096 ctx

    57.81 tok/s

    Decode

    312.83 tok/s

    Prefill

    2.17 s

    TTFT

    13.76 GB

    VRAM

    —

    MTP

    64.1 W

    Power draw

    L2 batch seed 2026-10-16: llama.cpp on hw_id=101 running model_id=60 at FP16