Arao AI
Back to search results

Qwen2.5-0.5B-Instruct · IQ4_XS

Start a discussion
NVIDIA RTX 509032GBllama.cpp v0.47L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen2.5-0.5B-Instruct (IQ4_XS) on NVIDIA RTX 5090 with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

319.07 tok/s

Decode

Decode speed

1,265.71 tok/s

Prefill

Prefill speed

0.22 s

TTFT

Time to first token

2.88 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 · IQ4_XS (current)Decode 319.07 · Prefill 1265.71 ·
  2. llama.cpp · Q8_0 Decode 358.97 · Prefill 1824.25 ·
  3. vLLM · IQ3_XS Decode 285.37 · Prefill 1069.59 ·

Core figures

Decode (avg)
319.07 tok/s
Prefill (avg)
1,265.71 tok/s
TTFT (avg)
0.22 s
VRAM (avg)
2.88 GB
MTP acceptance rate
—
TTFB
2.55 GB
Power draw
425.7 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen2.5-0.5B-Instruct
Quantization
IQ4_XS
Framework
llama.cpp
Version
v0.47
Context length
2048 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
NVIDIA RTX 5090
Nominal VRAM
32 GB
Measured VRAM (avg)
2.88 GB
OS
Ubuntu 22.04
Driver
555.85
CUDA
12.6
Power draw
425.7 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

    v0.47 · Ubuntu 22.04 · CUDA 12.6 · 2048 ctx

    319.07 tok/s

    Decode

    1,265.71 tok/s

    Prefill

    0.22 s

    TTFT

    2.88 GB

    VRAM

    —

    MTP

    425.7 W

    Power draw

    L2 batch seed 2026-10-16: llama.cpp on hw_id=13 running model_id=59 at IQ4_XS