Meme Generation on GOAT-G (test)
60.1BERTScore F1Gemini-3-Flash
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Gemini-3-Flashzero-shot=true2026.06 | 60.1 | 43.7 | |
| Claude-Sonnet-4.6zero-shot=true2026.06 | 58 | 41.2 | |
| Intent ProjectionBackbone=InternVL3-8B, Strategy=full2026.06 | 53.1 | 33.8 | |
| GPT-5.4-minizero-shot=true2026.06 | 52.4 | 33.8 | |
| Intent ProjectionBackbone=Qwen3-VL-8B, Strategy=full2026.06 | 52 | 32.3 | |
| InternVL3-8BStrategy=+ SFT + Chain + GRPOflat2026.06 | 51.2 | 31.9 | |
| Qwen3-VL-8BStrategy=+ SFT + Chain + GRPOflat2026.06 | 50.2 | 30.4 | |
| InternVL3-8BStrategy=+ SFT + Chain2026.06 | 49.6 | 30.3 | |
| Qwen3-VL-8BStrategy=+ SFT + Chain2026.06 | 48.6 | 28.9 | |
| InternVL3-8BStrategy=+ SFT2026.06 | 48.3 | 29.2 | |
| Intent ProjectionBackbone=LLaVA-OneVision-7B, Strategy=full2026.06 | 47.4 | 28.3 | |
| Qwen3-VL-8BStrategy=+ SFT2026.06 | 47.4 | 27.8 | |
| LLaVA-OneVision-7BStrategy=+ SFT + Chain + GRPOflat2026.06 | 45.6 | 26.6 | |
| InternVL3-8BStrategy=+ CoT2026.06 | 45.5 | 26 | |
| Qwen3-VL-8BStrategy=+ CoT2026.06 | 44 | 24.6 | |
| LLaVA-OneVision-7BStrategy=+ SFT + Chain2026.06 | 43.9 | 25.1 | |
| InternVL3-8BStrategy=Base2026.06 | 43.9 | 24.1 | |
| LLaVA-OneVision-7BStrategy=+ SFT2026.06 | 42.6 | 24 | |
| Qwen3-VL-8BStrategy=Base2026.06 | 42.1 | 22.5 | |
| LLaVA-OneVision-7BStrategy=+ CoT2026.06 | 39.4 | 20.8 | |
| LLaVA-OneVision-7BStrategy=Base2026.06 | 37.9 | 19.1 |