AgPerry/Video-R1-soft-filter-v2

---
license: apache-2.0
task_categories:
- visual-question-answering
- video-text-to-text
size_categories:
- 100K<n<1M
---

# Video-R1-soft-filter-v2

**147,850 samples** from Video-R1-260k filtered using the complete Kelsey soft filter.

## Filter Logic

Remove a sample only if **2+ models** can answer it text-only under circular evaluation.

| Model | Method | MCQ Coverage | Non-MCQ |
|-------|--------|-------------|---------|
| GPT-5-mini | Single-pass text-only | 263K (full) | ✅ |
| Qwen2.5-VL-7B | **4-perm circular eval** | 149,586 MCQ | pass@10 |
| Gemini 3.1 Pro | **3-perm circular eval** | 149,583 MCQ | direct eval (111K) |

## Comparison

| Version | Qwen | Gemini | Kept |
|---------|------|--------|------|
| soft-filter (v1) | 2-perm | Pro 3-perm | 160,837 |
| **soft-filter-v2** | **4-perm** | Pro 3-perm | **147,850** |

## Combination Breakdown

| GPT | Qwen | Gemini | Count | Action |
|-----|------|--------|-------|--------|
| N | N | N | 96,551 | **kept** |
| N | Y | N | 33,330 | **kept** |
| N | N | Y | 9,500 | **kept** |
| Y | N | N | 8,469 | **kept** |
| Y | Y | Y | 56,840 | removed |
| N | Y | Y | 42,329 | removed |
| Y | Y | N | 13,070 | removed |
| Y | N | Y | 2,982 | removed |

## Stats

- **Total**: 147,850 (56.2% of 260k)
- **Video**: computed from source data
- **Source**: Video-R1-260k (263,071 samples)