SauerkrautLM-Phi-3-medium: benchmark results
VAGOsolutions' German-English DPO tune of Microsoft's Phi-3-medium-4k-instruct (14B), fine-tuned with Spectrum QLoRA on 50% of layers (June 2024). Provider: VAGOsolutions. Released 2024-06-09. Access: Open.
Aggregate Rating 1453 ± 18, rank #980 of 1656 rated models, from 29 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard - MuSR | 48.45 | Score | 95.7 |
| Open Chinese LLM - GSM8K | 65.05 | Accuracy (%) | 89.6 |
| Open LLM Leaderboard - BBH | 64.33 | Score | 89.3 |
| Open Korean LLM - Ko-EQ-Bench | 37.28 | EQ-Bench score | 88.1 |
| Open Portuguese LLM - BLUEX | 66.2 | Accuracy (%) | 88 |
| Open Chinese LLM - ARC Challenge | 59.64 | Accuracy (%) | 85.7 |
| Open LLM Leaderboard - MMLU-Pro | 46.65 | Score | 85.6 |
| Open Portuguese LLM - ENEM | 73.41 | Accuracy (%) | 84.8 |
| Open LLM Leaderboard - GPQA | 33.47 | Score | 82.4 |
| Open Chinese LLM Leaderboard | 65.42 | Official Nonzero-Task Mean (%) | 82 |
| Open Portuguese LLM - ASSIN2 RTE | 92.68 | Macro F1 (%) | 82 |
| Open Korean LLM - Ko-GSM8K (flexible extract) | 45.11 | Exact match, flexible extract (%) | 80.1 |
Interactive version: theaggregate.ai/model?slug=sauerkrautlm-phi-3-medium · How It Works · Data refreshed daily, snapshot 2026-10-20.