Qwen3-1.7B

Qwen3-1.7B dense decoder language model for on-device text generation.

Qwen3-1.7B is Alibaba's 28-layer dense decoder (GQA attention, SwiGLU FFN). The hexagon package is a 1×HTP pin-all product path (refft-hexagon cli --model). Modeling is built from refft.hexagon.ops.

Model performance

Qwen3-1.7B on Hexagon HTP via refft-hexagon cli --model (1×HTP pin-all).

Dragonwing IQ-9075 (QCS9075) 1×HTP

SoC QCS9075 · Ubuntu 24.04.3 LTS (aarch64) · Hexagon v73 · w4 · runtime 0.6.5.dev682+g9833e0505 · measured 2026-08-13

Measured · QCS9075

38.5tok/s

TPS / Decode speed

88.8 ms

TTFT

668tok/s

Prefill speed

Install and run

Install

curl -fsSL https://raw.githubusercontent.com/refinefuture-ai/refft.cpp/main/refft-hexagon/install.sh | sh

Run

~/.local/share/refft-hexagon/bin/refft-hexagon cli --model ./Qwen3-1.7B-w4.refft --backend hexagon --prompt "Who are you?" --max_new_tokens 128

Serving, environment variables and full instructions

Qwen3-1.7B

Model performance

Install and run

Tags