SOLAR-10.7B-Instruct-truthy-dpo: benchmark results
A further fine-tune released by macadeliccc of SOLAR-10.7b-Instruct-dpo using three epochs of jondurbin/truthy-dpo-v0.1 after an earlier Intel/orca_dpo_pairs stage. The card's upstream version spelling is inconsistent, so no exact original SOLAR-version identity is inferred. Provider: macadeliccc. Access: Open.
Aggregate Rating 1457 ± 39, rank #3469 of 10916 rated models, from 6 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 76.75 | MC2 (%) (0-shot) | 96.3 |
| Open LLM Leaderboard v1 - HellaSwag | 88.44 | Normalized accuracy (%) (10-shot) | 91.8 |
| Open LLM Leaderboard v1 - ARC Challenge | 72.1 | Normalized accuracy (%) (25-shot) | 91 |
| Open LLM Leaderboard v1 - WinoGrande | 82.72 | Accuracy (%) (5-shot) | 85.2 |
| Open LLM Leaderboard v1 - MMLU | 65.45 | Accuracy (%) (5-shot) | 84.4 |
| Open LLM Leaderboard v1 - GSM8K | 59.21 | Accuracy (%) (5-shot) | 70.9 |
Interactive version: theaggregate.ai/model?slug=solar-10-7b-instruct-truthy-dpo · How It Works · Data refreshed daily, snapshot 2026-11-08.