Hateful Meme Classification and Explanation on FHM
77AccuracyL4-Scout
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| L4-ScoutPrompting Strategy=Standard, Backbone=Llama-4-Scout-17B2026.06 | 77 | 71 | 58 | 65.3 | 25.2 | |
| L4-ScoutPrompting Strategy=CoT, Backbone=Llama-4-Scout-17B2026.06 | 75.4 | 68.7 | 54.5 | 59.6 | 11.1 | |
| GPT-4.1Prompting Strategy=CoT2026.06 | 74.1 | 73.9 | 71.9 | 47 | 7.2 | |
| GPT-4.1Prompting Strategy=Standard2026.06 | 71.3 | 71.5 | 70 | 59.5 | 15.9 | |
| Kimi-TPrompting Strategy=CoT, Model Type=Thinking, Backbone=Kimi-VL-A3B2026.06 | 70.5 | 70.4 | 68.5 | 61.4 | 15.5 | |
| Qwen-TPrompting Strategy=CoT, Model Type=Thinking, Backbone=Qwen3-VL-8B2026.06 | 70.2 | 70.6 | 69.4 | 58.1 | 20.2 | |
| Fanar-2-Oryx-IVUPrompting Strategy=Standard2026.06 | 69 | 69 | 67 | 65 | 20 | |
| Fanar-2-Oryx-IVUPrompting Strategy=CoT2026.06 | 68 | 68 | 66 | 60 | 12 | |
| Kimi-IPrompting Strategy=CoT, Model Type=Instruct, Backbone=Kimi-VL-A3B2026.06 | 67.7 | 66.7 | 63.6 | 62.1 | 18.6 | |
| Kimi-TPrompting Strategy=Standard, Model Type=Thinking, Backbone=Kimi-VL-A3B2026.06 | 67.5 | 66.8 | 63.9 | 66 | 22.2 | |
| Kimi-IPrompting Strategy=Standard, Model Type=Instruct, Backbone=Kimi-VL-A3B2026.06 | 67 | 62.5 | 57.3 | 67.8 | 25.7 | |
| Qwen-TPrompting Strategy=Standard, Model Type=Thinking, Backbone=Qwen3-VL-8B2026.06 | 65.6 | 66.1 | 65.2 | 65.9 | 21.5 | |
| GemmaPrompting Strategy=CoT, Backbone=Gemma-3-12B2026.06 | 65.5 | 65.9 | 65.5 | 62.1 | 13.5 | |
| Qwen-IPrompting Strategy=CoT, Model Type=Instruct, Backbone=Qwen3-VL-8B2026.06 | 65.3 | 65.8 | 65.2 | 64.6 | 26.3 | |
| GemmaPrompting Strategy=Standard, Backbone=Gemma-3-12B2026.06 | 65.2 | 65.4 | 65.2 | 66.4 | 21.5 | |
| LlamaPrompting Strategy=Standard, Backbone=Llama-3.2-11B2026.06 | 64.2 | 59.9 | 54.7 | 66 | 23.4 | |
| Qwen-IPrompting Strategy=Standard, Model Type=Instruct, Backbone=Qwen3-VL-8B2026.06 | 63 | 63 | 63 | 66.2 | 21.6 | |
| LlamaPrompting Strategy=CoT, Backbone=Llama-3.2-11B2026.06 | 60 | 60.6 | 59.2 | 60 | 16.5 |