Meme Generation on MemeCap (test)
60.8BERTScore-F1Gemini-3-Flash
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Gemini-3-Flashzero-shot=true2026.06 | 60.8 | 43 | |
| Claude-Sonnet-4.6zero-shot=true2026.06 | 58.3 | 40.4 | |
| Intent ProjectionBackbone=InternVL3-8B, Strategy=full2026.06 | 54.5 | 33.1 | |
| Intent ProjectionBackbone=Qwen3-VL-8B, Strategy=full2026.06 | 53.7 | 32.4 | |
| InternVL3-8BStrategy=+ SFT + Chain + GRPOflat2026.06 | 52.4 | 31 | |
| GPT-5.4-minizero-shot=true2026.06 | 51.7 | 32.1 | |
| Qwen3-VL-8BStrategy=+ SFT + Chain + GRPOflat2026.06 | 51.5 | 30.2 | |
| InternVL3-8BStrategy=+ SFT + Chain2026.06 | 50.7 | 29.5 | |
| Qwen3-VL-8BStrategy=+ SFT + Chain2026.06 | 49.8 | 28.7 | |
| InternVL3-8BStrategy=+ SFT2026.06 | 49.5 | 28.4 | |
| Qwen3-VL-8BStrategy=+ SFT2026.06 | 48.6 | 27.5 | |
| Intent ProjectionBackbone=LLaVA-OneVision-7B, Strategy=full2026.06 | 47.8 | 27.5 | |
| InternVL3-8BStrategy=+ CoT2026.06 | 46.7 | 26.1 | |
| LLaVA-OneVision-7BStrategy=+ SFT + Chain + GRPOflat2026.06 | 45.8 | 25.7 | |
| Qwen3-VL-8BStrategy=+ CoT2026.06 | 45.2 | 24.6 | |
| InternVL3-8BStrategy=Base2026.06 | 45.1 | 24.3 | |
| LLaVA-OneVision-7BStrategy=+ SFT + Chain2026.06 | 44.1 | 24.3 | |
| Qwen3-VL-8BStrategy=Base2026.06 | 43.6 | 22.7 | |
| LLaVA-OneVision-7BStrategy=+ SFT2026.06 | 42.9 | 23.2 | |
| LLaVA-OneVision-7BStrategy=+ CoT2026.06 | 40.1 | 20.3 | |
| LLaVA-OneVision-7BStrategy=Base2026.06 | 38.4 | 18.2 |