Arao AI
Back to search results

Qwen3-Embedding-0.6B · FP16

Start a discussion
Apple M5 Max128GBvLLM v0.50L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3-Embedding-0.6B (FP16) on Apple M5 Max with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

257.66 tok/s

Decode

Decode speed

966.33 tok/s

Prefill

Prefill speed

1.43 s

TTFT

Time to first token

4.43 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 · Q6_K Decode 297.37 · Prefill 1423.38 ·
  2. vLLM · FP16 (current)Decode 257.66 · Prefill 966.33 ·
  3. vLLM · IQ2_M Decode 302.89 · Prefill 1094.91 ·
  4. Ollama · IQ3_XS Decode 290.71 · Prefill 1516.13 ·

Core figures

Decode (avg)
257.66 tok/s
Prefill (avg)
966.33 tok/s
TTFT (avg)
1.43 s
VRAM (avg)
4.43 GB
MTP acceptance rate
—
TTFB
2.06 GB
Power draw
186.8 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3-Embedding-0.6B
Quantization
FP16
Framework
vLLM
Version
v0.50
Context length
16384 tokens
Batch size
32
GPU layers
999
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)
4.43 GB
OS
Ubuntu 22.04
Driver
565.57
CUDA
12.4
Power draw
186.8 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

    v0.50 · Ubuntu 22.04 · CUDA 12.4 · 16384 ctx

    257.66 tok/s

    Decode

    966.33 tok/s

    Prefill

    1.43 s

    TTFT

    4.43 GB

    VRAM

    —

    MTP

    186.8 W

    Power draw

    L2 batch seed 2026-10-16: vLLM on hw_id=28 running model_id=62 at FP16