Chocolatine-14B-Instruct-4k-DPO: benchmark results
jpacifico's French-English DPO fine-tune of Phi-3-medium with a 4k context window, sibling of the 128k Chocolatine v1.2. Provider: jpacifico. Released 2024-08-01. Access: Open.
Aggregate Rating 1442 ± 21, rank #1043 of 1661 rated models, from 21 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Korean LLM - Ko-EQ-Bench | 44.08 | EQ-Bench score | 93.3 |
| Open LLM Leaderboard - BBH | 63 | Score | 86.8 |
| Open LLM Leaderboard - MMLU-Pro | 47.64 | Score | 86.5 |
| Open Portuguese LLM - ASSIN2 RTE | 93.09 | Macro F1 (%) | 86.3 |
| Open Korean LLM - Ko-IFEval | 41.65 | Strict accuracy, prompt/instruction mean (%) | 85 |
| Open LLM Leaderboard - GPQA | 34.14 | Score | 84.9 |
| Open Korean LLM - KorNAT-CKA | 30.41 | Normalized accuracy (%) | 83.2 |
| Open Portuguese LLM - ENEM | 72.22 | Accuracy (%) | 82.4 |
| Open Portuguese LLM - BLUEX | 62.17 | Accuracy (%) | 81.8 |
| Open Korean LLM - KorNAT-Harmless | 65.5 | Normalized accuracy (%) | 80.8 |
| Open Portuguese LLM - OAB Exams | 51.75 | Accuracy (%) | 80.8 |
| Open LLM Leaderboard - MuSR | 44.39 | Score | 80.3 |
Interactive version: theaggregate.ai/model?slug=chocolatine-14b-instruct-4k-dpo · How It Works · Data refreshed daily, snapshot 2026-10-23.