Arao AI
Back to search results

Qwen3-1.7B · IQ3_S

Start a discussion
Apple M1llama.cpp v0.28L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3-1.7B (IQ3_S) on Apple M1 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

97.72 tok/s

Decode

Decode speed

393.93 tok/s

Prefill

Prefill speed

1.89 s

TTFT

Time to first token

4.45 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 · IQ3_S (current)Decode 97.72 · Prefill 393.93 ·
  2. llama.cpp · Q5_K_M Decode 75.84 · Prefill 355.79 ·
  3. vLLM · Q4_K_M Decode 84.3 · Prefill 383.2 ·
  4. vLLM · Q6_K Decode 72.53 · Prefill 321.64 ·

Core figures

Decode (avg)
97.72 tok/s
Prefill (avg)
393.93 tok/s
TTFT (avg)
1.89 s
VRAM (avg)
4.45 GB
MTP acceptance rate
—
TTFB
0.78 GB
Power draw
62.7 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3-1.7B
Quantization
IQ3_S
Framework
llama.cpp
Version
v0.28
Context length
32768 tokens
Batch size
256
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
Measured VRAM (avg)
4.45 GB
OS
macOS 15.0
Driver
Apple M1
CUDA
N/A
Power draw
62.7 W

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-16

    v0.28 · macOS 15.0 · CUDA N/A · 32768 ctx

    97.72 tok/s

    Decode

    393.93 tok/s

    Prefill

    1.89 s

    TTFT

    4.45 GB

    VRAM

    —

    MTP

    62.7 W

    Power draw

    L2 batch seed 2026-10-16: llama.cpp on hw_id=99 running model_id=55 at IQ3_S