Arao AI
Back to search results

Qwen3.5-4B · Q3_K_M

Start a discussion
NVIDIA RTX 507012GBvLLM v0.29L2 Cross-framework verified

This page aggregates 1 real-world runs of Qwen3.5-4B (Q3_K_M) on NVIDIA RTX 5070 with vLLM, contributed by 1 independent source platforms; metrics are averages of published measurements.

112.72 tok/s

Decode

Decode speed

436.5 tok/s

Prefill

Prefill speed

2.21 s

TTFT

Time to first token

7.81 GB

VRAM

VRAM usage

L2 Cross-framework verified

1

Measured runs

1

Independent sources

Other

Source platforms

Today

Last verified

Performance

  1. vLLM · Q3_K_M (current)Decode 112.72 · Prefill 436.5 ·
  2. vLLM · Q4_K_M Decode 84.51 · Prefill 408.98 ·
  3. Strata · IQ3_S Decode 154.26 · Prefill 563.04 ·
  4. llama.cpp · Q5_K_M Decode 174.7 · Prefill 1310.3 ·

Core figures

Decode (avg)
112.72 tok/s
Prefill (avg)
436.5 tok/s
TTFT (avg)
2.21 s
VRAM (avg)
7.81 GB
MTP acceptance rate
—
TTFB
0.9 GB
Power draw
229.1 W

Configuration

Member-level fields are taken from the most recent run

Model
Qwen3.5-4B
Quantization
Q3_K_M
Framework
vLLM
Version
v0.29
Context length
8192 tokens
Batch size
16
GPU layers
999
Flash Attention
On

Hardware

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

GPU
NVIDIA RTX 5070
Nominal VRAM
12 GB
Measured VRAM (avg)
7.81 GB
OS
macOS 15.0
Driver
555.85
CUDA
12.4
Power draw
229.1 W

Sources & evidence

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

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

    v0.29 · macOS 15.0 · CUDA 12.4 · 8192 ctx

    112.72 tok/s

    Decode

    436.5 tok/s

    Prefill

    2.21 s

    TTFT

    7.81 GB

    VRAM

    —

    MTP

    229.1 W

    Power draw

    L2 batch seed 2026-10-10: vLLM on hw_id=18 running model_id=48 at Q3_K_M