---
license: apache-2.0
tags:
  - vision-language
  - multimodal
  - visual-reasoning
  - self-evolving
  - self-distillation
---

<div align="center">

# SelfThinker Data: Reliability-Verified Visual Thought Pools & Benchmark Suite

[![Models](https://img.shields.io/badge/%F0%9F%A4%97%20Models-SelfThinker--7B-FFB000)](https://huggingface.co/jankin123/SelfThinker-7B)
[![Data](https://img.shields.io/badge/%F0%9F%A4%97%20Data-SelfThinker-0072B2)](https://huggingface.co/jankin123/SelfThinker)
[![Code](https://img.shields.io/badge/GitHub-SelfThinker-181717?logo=github)](https://github.com/zhangquanchen/SelfThinker)

</div>

This repository contains the reliability-verified intrinsic visual-thought training sets and the benchmark TSV files for training and evaluating **SelfThinker**.

- **Models:** [https://huggingface.co/jankin123/SelfThinker-7B](https://huggingface.co/jankin123/SelfThinker-7B)
- **Data:** [https://huggingface.co/jankin123/SelfThinker](https://huggingface.co/jankin123/SelfThinker)
- **Code:** [https://github.com/zhangquanchen/SelfThinker](https://github.com/zhangquanchen/SelfThinker)

<p align="center">
  <img src="assets/Fig5.png" width="100%" alt="Reliability-Verified Visual Thoughts on Navigation and Visual Search"/>
</p>

---

## 1. Reliability-Verified Training Sets

Synthesized multi-object visual-search problems (`384x384` px canvases with 5–8 non-overlapping shapes across 5 shapes and 8 colors) where candidate intrinsic visual thoughts $\{v^{(k)}\}_{k=1}^{6}$ are scored by conditional pointwise mutual information:

$$u(v) = \log p_\theta\big(d \mid C(I, q, v)\big) - \log p_\theta\big(d \mid C(I, q)\big)$$

and retained when the top candidate $v^\star$ exceeds the item-specific 95th-percentile mismatched threshold $Q_{0.95}(s_i)$.

| Directory | Verification Direction $d$ | Verified Items | Parquet File | Command |
|---|---|---:|---|---|
| `audit_pass_l0_vsearch_trainonly/` | Ground-truth answer $a^\star$ | 1,092 / 1,500 | `Visual_Search/data/train-00000-of-00001.parquet` (111.7 MB) | `scripts/train_selfthinker.sh answer` |
| `audit_pass_l1_vsearch_trainonly/` | Label-free rollout majority vote $\tilde{a}$ | 1,074 / 1,500 | `Visual_Search/data/train-00000-of-00001.parquet` (107.4 MB) | `scripts/train_selfthinker.sh vote` |

<p align="center">
  <img src="assets/Fig7.png" width="100%" alt="Answer Consensus vs Visual Consensus on Visual Search"/>
</p>

---

## 2. Benchmark Evaluation Suite (`lmudata/`)

`lmudata/` provides all self-contained TSV files (with embedded base64 images) used by the seven-benchmark evaluation script (`scripts/eval_nine.sh`) and the multi-layer image-condition probe (`scripts/eval_image.sh`):

| Benchmark | TSV File(s) | Items | Size | Aggregation Rule |
|---|---|---:|---:|---|
| **VSP** | `lmudata/VSP_maze_task_main_original.tsv` | 600 | 144.3 MB | Main maze task accuracy |
| **ChartQA** | `lmudata/ChartQA_h_bar.tsv`, `lmudata/ChartQA_v_bar.tsv` | 444 + 382 | 35.7 MB, 19.7 MB | Mean of horizontal and vertical bar subsets |
| **V\*** | `lmudata/VStarBench.tsv` | 191 | 162.0 MB | Direct accuracy |
| **MMVP** | `lmudata/MMVP.tsv` | 300 | 4.2 MB | Direct accuracy |
| **SAT** | `lmudata/SAT_circular.tsv` | 150 | 354.1 MB | Circular evaluation split accuracy |
| **BLINK** | `lmudata/BLINK.tsv` | 1,901 | 538.4 MB | Overall accuracy across 14 subtasks |
| **CV-Bench** | `lmudata/CV-Bench-2D.tsv`, `lmudata/CV-Bench-3D.tsv` | 1,438 + 1,200 | 139.9 MB, 158.2 MB | Mean of 2D and 3D subsets |

<p align="center">
  <img src="assets/Fig10.png" width="100%" alt="Conditions of the Multi-Layer Image-Condition Probe"/>
</p>

---

## Download & Usage

```bash
git clone https://github.com/zhangquanchen/SelfThinker.git
cd SelfThinker

huggingface-cli download jankin123/SelfThinker --local-dir data
huggingface-cli download jankin123/SelfThinker-7B --local-dir models
```