---
license: apache-2.0
base_model: Qwen/Qwen3-VL-2B-Instruct
library_name: transformers.js
pipeline_tag: image-text-to-text
tags:
  - onnx
  - webgpu
  - interpretability
  - attention-steering
---

# Qwen3-VL-2B-Instruct — Gaze-Head Steering (ONNX, WebGPU)

ONNX build of [Qwen3-VL-2B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-2B-Instruct)
(via [onnx-community's export](https://huggingface.co/onnx-community/Qwen3-VL-2B-Instruct-ONNX))
with two extra decoder inputs that expose the **gaze-head steering** intervention
from the paper *Gaze Heads: How VLMs Look at What They Describe*:

- `gaze_head_mask` `(28, 16)` float16 — 1.0 selects an attention head
- `gaze_sign` `(total_sequence_length,)` float16 — additive pre-softmax
  attention bias per KV position (e.g. +10000 on target image tokens,
  -10000 on other image tokens, 0 on text)

Feeding zeros reproduces the original model. The layers hosting the top-20
gaze heads use decomposed attention (plain ops) so the bias works on the
onnxruntime-web WebGPU backend, whose fused GroupQueryAttention kernel
ignores `attention_bias`; all other layers keep the fused kernel.

`vision_encoder_q4f16.onnx` is a **stub**: the demo injects precomputed
image features and never runs the vision tower. Use the onnx-community
repo if you need the real encoder.

Only the decoder graph file differs from the upstream export; all weight
data files are bit-identical copies.

**Demo / paper / code:** see the project page.