Qwen2.5-7B-Instruct-abliterated-v2: benchmark results
huihui-ai's abliterated variant of Qwen2.5 7B Instruct with refusal directions removed, a v2 refinement of their uncensored release. Provider: huihui-ai. Released 2024-09-22. Access: Open.
Aggregate Rating 1535 ± 17, rank #659 of 1656 rated models, from 144 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Portuguese LLM - FaQuAD NLI | 85.45 | Macro F1 (%) | 98.7 |
| Open Korean LLM - Ko-IFEval | 63.72 | Strict accuracy, prompt/instruction mean (%) | 97.8 |
| Open Ko-LLM Leaderboard | 49.87 | Average (%) | 97.1 |
| Open Korean LLM Leaderboard | 49.87 | Average Score (%) | 97.1 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 68.23 | Exact match, flexible extract (%) | 96.9 |
| Open Korean LLM - Ko-EQ-Bench | 55.82 | EQ-Bench score | 96.8 |
| Open LLM Leaderboard - MATH Level 5 | 46.37 | Score | 94.4 |
| Open Korean LLM - KorNAT-Helpful | 53.37 | Normalized accuracy (%) | 94.3 |
| Open Portuguese LLM - ASSIN2 RTE | 93.83 | Macro F1 (%) | 92.4 |
| Open LLM Leaderboard - IFEval | 76.06 | Score | 91.6 |
| Open Korean LLM - KorNAT-Harmless | 67.15 | Normalized accuracy (%) | 88.2 |
| Open Korean LLM - KorNAT-CKA | 30.96 | Normalized accuracy (%) | 85.6 |
Interactive version: theaggregate.ai/model?slug=qwen2-5-7b-instruct-abliterated-v2 · How It Works · Data refreshed daily, snapshot 2026-10-20.