Arao AI
Back to search results

Qwen3.5-4B · IQ2_M

Start a discussion
Apple M1llama.cpp v0.50L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.5-4B (IQ2_M) on Apple M1 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

88.05 tok/s

Decode

Decode speed

311.52 tok/s

Prefill

Prefill speed

2.24 s

TTFT

Time to first token

6.47 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 · IQ2_M (current)Decode 88.05 · Prefill 311.52 ·
  2. llama.cpp · Q4_K_M Decode 55.57 · Prefill 194.52 ·
  3. Strata · Q4_K_S Decode 87.2 · Prefill 469.35 ·

Core figures

Decode (avg)
88.05 tok/s
Prefill (avg)
311.52 tok/s
TTFT (avg)
2.24 s
VRAM (avg)
6.47 GB
MTP acceptance rate
—
TTFB
0.7 GB
Power draw
45.7 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.5-4B
Quantization
IQ2_M
Framework
llama.cpp
Version
v0.50
Context length
2048 tokens
Batch size
512
GPU layers
99
Flash Attention
Off

Hardware

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

GPU
Apple M1
Measured VRAM (avg)
6.47 GB
OS
macOS 15.0
Driver
Apple M1
CUDA
N/A
Power draw
45.7 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

    v0.50 · macOS 15.0 · CUDA N/A · 2048 ctx

    88.05 tok/s

    Decode

    311.52 tok/s

    Prefill

    2.24 s

    TTFT

    6.47 GB

    VRAM

    —

    MTP

    45.7 W

    Power draw

    L2 batch seed 2026-10-16: llama.cpp on hw_id=99 running model_id=48 at IQ2_M