Multimodal idiomatic understanding on Varnika (test)
0.54ROUGE-1 ScoreQwen2.5-VL-7B-Instruct
Evaluation Results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Qwen2.5-VL-7B-InstructExperimental Setting=VLM+HybridMoE2026.06 | 0.54 | 0.32 | 0.41 | 0.43 | 56.32 | 28.92 | 21.48 | 24.19 | 0.95 | 0.82 | 0.47 | |
| Qwen2.5-VL-7B-InstructExperimental Setting=VLM+Finetune2026.06 | 0.48 | 0.22 | 0.36 | 0.37 | 51.06 | 24.81 | 16.51 | 20.68 | 0.93 | 0.75 | 0.46 | |
| SmolVLM-InstructExperimental Setting=VLM+HybridMoE2026.06 | 0.48 | 0.21 | 0.36 | 0.36 | 50.46 | 24.16 | 16.59 | 20.92 | 0.92 | 0.66 | 0.35 | |
| Gemma3-12b-ptExperimental Setting=VLM+HybridMoE2026.06 | 0.46 | 0.19 | 0.35 | 0.35 | 38.81 | 17.82 | 11.93 | 14.58 | 0.92 | 0.79 | 0.55 | |
| Paligemma2-10BExperimental Setting=VLM+HybridMoE2026.06 | 0.45 | 0.2 | 0.34 | 0.35 | 35.97 | 17.25 | 14.2 | 15.94 | 0.89 | 0.59 | 0.32 | |
| SmolVLM-InstructExperimental Setting=VLM+Finetune2026.06 | 0.43 | 0.18 | 0.33 | 0.33 | 47.48 | 21.82 | 14.06 | 17.7 | 0.92 | 0.59 | 0.35 | |
| Gemma3-12b-ptExperimental Setting=VLM+Finetune2026.06 | 0.4 | 0.15 | 0.29 | 0.29 | 32.86 | 13.36 | 7.9 | 11.64 | 0.9 | 0.74 | 0.54 | |
| Llava-1.5-7BExperimental Setting=VLM+HybridMoE2026.06 | 0.4 | 0.17 | 0.32 | 0.34 | 32.52 | 16.64 | 12.95 | 14.25 | 0.91 | 0.61 | 0.33 | |
| Paligemma2-10BExperimental Setting=VLM+Finetune2026.06 | 0.39 | 0.15 | 0.27 | 0.29 | 32.62 | 15.73 | 8.95 | 13.73 | 0.88 | 0.53 | 0.32 | |
| Llava-1.5-7BExperimental Setting=VLM+Finetune2026.06 | 0.35 | 0.15 | 0.26 | 0.26 | 27.46 | 12.74 | 8.2 | 11.4 | 0.9 | 0.57 | 0.31 | |
| Blip2-7BExperimental Setting=VLM+HybridMoE2026.06 | 0.3 | 0.13 | 0.23 | 0.28 | 23.76 | 12.93 | 13.29 | 15.64 | 0.89 | 0.55 | 0.33 | |
| Blip2-7BExperimental Setting=VLM+Finetune2026.06 | 0.26 | 0.07 | 0.17 | 0.18 | 18.64 | 8.62 | 6.28 | 9.17 | 0.88 | 0.52 | 0.32 |