zephyr-220m-dpo-full: benchmark results
BEE-spoke-data release: A derivative of amazingvince/zephyr-220m-sft-full whose publisher card lists ultrafeedback_binarized and preference-reward training metrics. It remains distinct from the supervised fine-tune and Hugging Face's larger Zephyr releases. Provider: BEE-spoke-data. Access: Open.
Aggregate Rating 1110 ± 40, rank #10609 of 10916 rated models, from 6 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 43.44 | MC2 (%) (0-shot) | 23.5 |
| Open LLM Leaderboard v1 - GSM8K | 0.53 | Accuracy (%) (5-shot) | 12.7 |
| Open LLM Leaderboard v1 - MMLU | 26.43 | Accuracy (%) (5-shot) | 12.5 |
| Open LLM Leaderboard v1 - HellaSwag | 29.15 | Normalized accuracy (%) (10-shot) | 5.1 |
| Open LLM Leaderboard v1 - ARC Challenge | 25.43 | Normalized accuracy (%) (25-shot) | 4.7 |
| Open LLM Leaderboard v1 - WinoGrande | 50.99 | Accuracy (%) (5-shot) | 4.7 |
Interactive version: theaggregate.ai/model?slug=zephyr-220m-dpo-full · How It Works · Data refreshed daily, snapshot 2026-11-08.