Arao AI
Back to search results

Hy-Memory-30B-A3B · IQ3_XS

Start a discussion
Apple M5 Max128GBllama.cpp v0.26L2 Cross-framework verified

This page aggregates 1 real-world runs of Hy-Memory-30B-A3B (IQ3_XS) on Apple M5 Max with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

147.71 tok/s

Decode

Decode speed

746.12 tok/s

Prefill

Prefill speed

0.53 s

TTFT

Time to first token

41.16 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

16 days ago

Last verified

Performance

  1. llama.cpp · IQ3_XS (current)Decode 147.71 · Prefill 746.12 ·
  2. llama.cpp · Q3_K_M Decode 134.94 · Prefill 533.49 ·
  3. Ollama · IQ2_XS Decode 139.95 · Prefill 497.06 ·
  4. Ollama · Q8_0 Decode 63.92 · Prefill 252.01 ·

Core figures

Decode (avg)
147.71 tok/s
Prefill (avg)
746.12 tok/s
TTFT (avg)
0.53 s
VRAM (avg)
41.16 GB
MTP acceptance rate
—
TTFB
1.18 GB
Power draw
68.7 W

Configuration

Member-level fields are taken from the most recent run

Model
Hy-Memory-30B-A3B
Quantization
IQ3_XS
Framework
llama.cpp
Version
v0.26
Context length
4096 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 M5 Max
Nominal VRAM
128 GB
Measured VRAM (avg)
41.16 GB
OS
macOS 26.1
Power draw
68.7 W

Get started

Original model resources

Base model resources may not match this quantization. Use the verified deployment weights above when available.

Sources & evidence

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

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

    v0.26 · macOS 26.1 · 4096 ctx

    147.71 tok/s

    Decode

    746.12 tok/s

    Prefill

    0.53 s

    TTFT

    41.16 GB

    VRAM

    —

    MTP

    68.7 W

    Power draw

    L2 batch seed 2026-10-19: llama.cpp on hw_id=28 running model_id=87 at IQ3_XS