ALMA-13B-Pretrain: benchmark results
haoranxu's ALMA-13B-Pretrain, the monolingual-data fine-tuning stage of the ALMA translation model built on Llama 2 13B. Provider: Johns Hopkins University. Released 2023-09-17. Access: Open.
Aggregate Rating 1299 ± 24, rank #1457 of 1635 rated models, from 19 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - WinoGrande | 76.4 | Accuracy (%) (5-shot) | 48 |
| Open LLM Leaderboard v1 - HellaSwag | 80.15 | Normalized accuracy (%) (10-shot) | 40.3 |
| Open LLM Leaderboard v1 - ARC Challenge | 56.91 | Normalized accuracy (%) (25-shot) | 36.3 |
| Open Portuguese LLM - ENEM | 45.7 | Accuracy (%) | 31.9 |
| Open Chinese LLM - WinoGrande | 61.8 | Accuracy (%) | 31.8 |
| Open Portuguese LLM - BLUEX | 37.27 | Accuracy (%) | 30.6 |
| Open Portuguese LLM - OAB Exams | 33.17 | Accuracy (%) | 30 |
| Open Portuguese LLM - ASSIN2 RTE | 78.75 | Macro F1 (%) | 29.7 |
| Open LLM Leaderboard v1 - MMLU | 50.31 | Accuracy (%) (5-shot) | 29.6 |
| Open LLM Leaderboard v1 - GSM8K | 8.87 | Accuracy (%) (5-shot) | 29.2 |
| Open Chinese LLM - HellaSwag | 54.47 | Accuracy (%) | 21.3 |
| Open Chinese LLM - ARC Challenge | 44.62 | Accuracy (%) | 14.8 |
Interactive version: theaggregate.ai/model?slug=alma-13b-pretrain · How It Works · Data refreshed daily, snapshot 2026-11-05.