Arao AI
Back to search results

Qwen2.5-Omni-3B · Q3_K_L

Start a discussion
NVIDIA Jetson AGX Orin 64GB64GBllama.cpp v0.31L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen2.5-Omni-3B (Q3_K_L) on NVIDIA Jetson AGX Orin 64GB with llama.cpp, contributed by 1 independent source platforms; metrics are averages of published measurements.

91.94 tok/s

Decode

Decode speed

356.4 tok/s

Prefill

Prefill speed

0.48 s

TTFT

Time to first token

6.64 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

5 days ago

Last verified

Performance

  1. vLLM · IQ1_S Decode 135.17 · Prefill 565.39 ·
  2. vLLM · Q2_K Decode 118.95 · Prefill 465.31 ·
  3. Strata · Q3_K_S Decode 147.6 · Prefill 732.7 ·
  4. llama.cpp · Q2_K Decode 113.33 · Prefill 517.85 ·
  5. llama.cpp · Q3_K_L (current)Decode 91.94 · Prefill 356.4 ·
  6. vLLM · Q4_K_S Decode 88.52 · Prefill 343.83 ·

Core figures

Decode (avg)
91.94 tok/s
Prefill (avg)
356.4 tok/s
TTFT (avg)
0.48 s
VRAM (avg)
6.64 GB
MTP acceptance rate
—
TTFB
0.74 GB
Power draw
42.6 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen2.5-Omni-3B
Quantization
Q3_K_L
Framework
llama.cpp
Version
v0.31
Context length
4096 tokens
Batch size
256
GPU layers
99
Flash Attention
On

Hardware

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

GPU
NVIDIA Jetson AGX Orin 64GB
Nominal VRAM
64 GB
Measured VRAM (avg)
6.64 GB
OS
JetPack 6.0
Driver
NVIDIA L4T 36.3
CUDA
12.6
Power draw
42.6 W

Sources & evidence

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

  1. L2 Cross-framework verifiedOtherOriginal link Verified on 2026-10-13

    v0.31 · JetPack 6.0 · CUDA 12.6 · 4096 ctx

    91.94 tok/s

    Decode

    356.4 tok/s

    Prefill

    0.48 s

    TTFT

    6.64 GB

    VRAM

    —

    MTP

    42.6 W

    Power draw

    L2 batch seed 2026-10-13: llama.cpp on hw_id=76 running model_id=60 at Q3_K_L