Llama-3.1-8B-Instruct-abliterated_via_adapter: benchmark results
grimjim's Llama 3.1 8B Instruct with refusals removed by task-arithmetic-merging an abliteration LoRA extracted from failspy's Llama 3 8B abliterated-v3. Provider: grimjim. Released 2024-07-25. Access: Open.
Aggregate Rating 1470 ± 19, rank #893 of 1661 rated models, from 21 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM Leaderboard | 44.63 | Average Score (%) | 90.7 |
| Open Ko-LLM Leaderboard | 44.63 | Average (%) | 90.6 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 55.65 | Exact match, flexible extract (%) | 89.7 |
| Open Korean LLM - Ko-IFEval | 44.31 | Strict accuracy, prompt/instruction mean (%) | 88.7 |
| Open Korean LLM - Ko-GPQA-Diamond | 26.26 | Normalized accuracy (%) | 85.5 |
| Open Korean LLM - Ko-EQ-Bench | 31.28 | EQ-Bench score | 80.4 |
| Open Portuguese LLM - OAB Exams | 51.44 | Accuracy (%) | 80 |
| Open Portuguese LLM - BLUEX | 60.22 | Accuracy (%) | 78 |
| Open Korean LLM - KorNAT-CKA | 28.71 | Normalized accuracy (%) | 76.9 |
| Open Portuguese LLM - ENEM | 70.54 | Accuracy (%) | 76.7 |
| Open LLM Leaderboard - GPQA | 31.38 | Score | 70.7 |
| Open Portuguese LLM - ASSIN2 RTE | 91.69 | Macro F1 (%) | 69.3 |
Interactive version: theaggregate.ai/model?slug=llama-3-1-8b-instruct-abliterated-via-adapter · How It Works · Data refreshed daily, snapshot 2026-10-23.