Hate speech classification on Implicit Hate (test)
0.67Macro F1Llama_Full
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Llama_FullTraining Paradigm=Full, Model Backbone=Llama, Data%=100%2025.09 | 0.67 | 100 | |
| T5_FullTraining Paradigm=Full, Model Backbone=T5, Data%=100%2025.09 | 0.64 | 96 | |
| ModernBERTTraining Paradigm=Full, Model Backbone=ModernBERT, Data%=100%2025.09 | 0.64 | 96 | |
| SMARTER (Llama_DPO-256)Training Paradigm=DPO, K shots=256, Model Backbone=Llama, Data%=57%2025.09 | 0.6 | 90 | |
| SMARTER (T5_DPO-256)Training Paradigm=DPO, K shots=256, Model Backbone=T5, Data%=57%2025.09 | 0.59 | 88 | |
| GPT-5-chatTraining Paradigm=Zero-shot, K shots=0, Model Backbone=GPT-5-chat2025.09 | 0.58 | 87 | |
| GPT-4.1Training Paradigm=Zero-shot, K shots=0, Model Backbone=GPT-4.12025.09 | 0.57 | 85 | |
| Qwen-32BTraining Paradigm=Zero-shot, K shots=0, Model Backbone=Qwen-32B2025.09 | 0.49 | 73 | |
| Qwen-32BTraining Paradigm=16-shot ICL, K shots=16, Model Backbone=Qwen-32B2025.09 | 0.48 | 72 | |
| GPT-4o-miniTraining Paradigm=Zero-shot, K shots=0, Model Backbone=GPT-4o-mini2025.09 | 0.42 | 63 | |
| GPT-5-chatTraining Paradigm=16-shot ICL, K shots=16, Model Backbone=GPT-5-chat2025.09 | 0.4 | 60 | |
| GPT-4.1Training Paradigm=16-shot ICL, K shots=16, Model Backbone=GPT-4.12025.09 | 0.38 | 57 | |
| GPT-4o-miniTraining Paradigm=16-shot ICL, K shots=16, Model Backbone=GPT-4o-mini2025.09 | 0.15 | 22 |