| ================================================================================ | |
| DETOXIFY-MEDIUM MODEL BENCHMARK RESULTS | |
| ================================================================================ | |
| 📊 EXECUTIVE SUMMARY | |
| -------------------------------------------------- | |
| Benchmark Date: 2025-09-18 13:17:58 | |
| Model: Detoxify-Medium | |
| Dataset: ParaDetox (https://github.com/s-nlp/paradetox) | |
| Total Samples: 1011 | |
| 🎯 OVERALL PERFORMANCE METRICS | |
| -------------------------------------------------- | |
| Toxicity Reduction: 0.178 | |
| Semantic Preservation: 0.561 | |
| Fluency: 0.929 | |
| Average Latency: 160.2ms | |
| Original Toxicity: 0.196 | |
| Final Toxicity: 0.018 | |
| 📈 DATASET BREAKDOWN | |
| -------------------------------------------------- | |
| 🔹 PARADETOX TOXIC NEUTRAL | |
| Samples: 1000 | |
| Toxicity Reduction: 0.044 | |
| Semantic Preservation: 0.645 | |
| Fluency: 0.922 | |
| Latency: 156.2ms | |
| Original Toxicity: 0.051 | |
| Final Toxicity: 0.007 | |
| 🔹 PARADETOX HIGH TOXICITY | |
| Samples: 11 | |
| Toxicity Reduction: 0.313 | |
| Semantic Preservation: 0.477 | |
| Fluency: 0.936 | |
| Latency: 164.3ms | |
| Original Toxicity: 0.342 | |
| Final Toxicity: 0.029 | |
| ================================================================================ |