Privacy-preserving dialogue processing on Med-PCD (test)
91.5QualityGPT-4o-mini
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| GPT-4o-miniVariant=Unredacted2025.10 | 91.5 | 100 | — | — | |
| Privacy-R1Backbone=Llama-3.1-8B-Instruct2025.10 | 89.5 | 5.1 | 7.5 | 4.1 | |
| Privacy-R1Backbone=Qwen2-7B-Instruct2025.10 | 88.4 | 12 | 12.2 | 6.5 | |
| Privacy-R1Backbone=Mistral-7B-Instruct-v0.32025.10 | 87.9 | 9.5 | 13.4 | 4.5 | |
| PAPILLONBackbone=Llama-3.1-8B-Instruct2025.10 | 82 | 9.2 | — | — | |
| Privacy-R1Backbone=Llama-3.2-3B-Instruct2025.10 | 81 | 15.4 | 22.5 | 12.7 | |
| PAPILLONBackbone=Qwen2-7B-Instruct2025.10 | 76.2 | 18.5 | — | — | |
| Privacy-R1Backbone=Llama-3.2-1B-Instruct2025.10 | 75.3 | 18.2 | 30.2 | 24.3 | |
| PAPILLONBackbone=Mistral-7B-Instruct-v0.32025.10 | 74.5 | 14 | — | — | |
| GPT-4o-miniVariant=Redacted2025.10 | 72.3 | 0 | — | — | |
| Heuristic Router (SFT-only)Backbone=Qwen2-7B-Instruct2025.10 | 71 | 3 | — | — | |
| Heuristic Router (SFT-only)Backbone=Llama-3.1-8B-Instruct2025.10 | 70.5 | 2.5 | — | — | |
| Heuristic Router (SFT-only)Backbone=Mistral-7B-Instruct-v0.32025.10 | 69.1 | 2.8 | — | — | |
| Always-LocalBackbone=Qwen2-7B-Instruct2025.10 | 60.1 | 0 | — | — | |
| PAPILLONBackbone=Llama-3.2-3B-Instruct2025.10 | 58.5 | 28.1 | — | — | |
| Always-LocalBackbone=Mistral-7B-Instruct-v0.32025.10 | 58.3 | 0 | — | — | |
| Heuristic Router (SFT-only)Backbone=Llama-3.2-3B-Instruct2025.10 | 55.6 | 3.1 | — | — | |
| Always-LocalBackbone=Llama-3.1-8B-Instruct2025.10 | 55 | 0 | — | — | |
| PAPILLONBackbone=Llama-3.2-1B-Instruct2025.10 | 45.1 | 42.5 | — | — | |
| Heuristic Router (SFT-only)Backbone=Llama-3.2-1B-Instruct2025.10 | 40.8 | 3.5 | — | — | |
| Always-LocalBackbone=Llama-3.2-3B-Instruct2025.10 | 40.1 | 0 | — | — | |
| Always-LocalBackbone=Llama-3.2-1B-Instruct2025.10 | 25.5 | 0 | — | — |