Llama 3 70B Japanese Suzume Vector V0.1: benchmark results
mmnga's experimental Japanese adaptation of Llama 3 70B Instruct, applying an upscaled chat vector from lightblue's Suzume 8B Japanese fine-tune. Provider: mmnga. Released 2024-04-28. Access: Open.
Aggregate Rating 1506 ± 22, rank #765 of 1656 rated models, from 20 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - GSM8K | 87.41 | Accuracy (%) (5-shot) | 100 |
| Open LLM Leaderboard v1 - MMLU | 80.28 | Accuracy (%) (5-shot) | 99.7 |
| Open Chinese LLM - GSM8K | 70.58 | Accuracy (%) | 97.3 |
| Open Chinese LLM - CMMLU | 72.14 | Accuracy (%) | 95 |
| Open LLM Leaderboard - BBH | 65.42 | Score | 92.4 |
| Open LLM Leaderboard v1 - ARC Challenge | 72.35 | Normalized accuracy (%) (25-shot) | 91.7 |
| Open LLM Leaderboard - MMLU-Pro | 52.24 | Score | 90.6 |
| Open LLM Leaderboard v1 - WinoGrande | 82.79 | Accuracy (%) (5-shot) | 85.6 |
| Open Chinese LLM - ARC Challenge | 59.56 | Accuracy (%) | 85.2 |
| Open Chinese LLM Leaderboard | 66.23 | Official Nonzero-Task Mean (%) | 83.4 |
| Open Chinese LLM - C-Eval Semantic | 86.76 | Accuracy (%) | 81.8 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 62.93 | MC2 (%) (0-shot) | 79.9 |
Interactive version: theaggregate.ai/model?slug=llama-3-70b-japanese-suzume-vector-v0-1 · How It Works · Data refreshed daily, snapshot 2026-10-20.