DGrid AI (@dgrid_ai) “2/📊 Key Results (held-out test set, n=300) ✅DeBERTa Judge: Pearson 0.747 (95%” — TopicDigg
DGrid AI
@dgrid_ai
💡 DGrid AI is a Decentralized AI Smart Network for Cost-effective, Reliable & Verifiable AI. One API, ALL LLMs. 🌟TG:
加入 March 2024
103 正在关注 132.7K 粉丝
2/📊 Key Results (held-out test set, n=300)
✅DeBERTa Judge: Pearson 0.747 (95% CI [0.663, 0.816]) → Outperforms all reference-based evaluators in our prior framework (best: 0.629)
✅Reference-Free composite score: Pearson 0.645→ Matches the best reference-based single evaluator — with zero reference answers
✅Cascade + online weight calibration: Saves 72.7% evaluation cost
显示更多
