Query Accuracy on ADE20K Fresh 500 examples (test)
100Fence AccuracyRegex Baseline
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Regex BaselineSetting=base, Type=Non-learning baseline2026.06 | 100 | 100 | 100 | 74.4 | 5.6 | |
| GLAREBackbone=Gemma 2 2B, Setting=Fine-tuned, Tuning=QLoRA2026.06 | 100 | 100 | 100 | 95.4 | 2.8 | |
| Gemma 2 9BSetting=base2026.06 | 100 | 3.2 | 3.2 | 0.2 | 0.4 | |
| GLAREBackbone=Gemma 2 9B, Setting=Fine-tuned, Tuning=QLoRA2026.06 | 100 | 100 | 100 | 95.2 | 2.4 | |
| Gemma 2 27BSetting=base2026.06 | 100 | 0.6 | 0.6 | 0 | 0.2 | |
| GLAREBackbone=Gemma 2 27B, Setting=Fine-tuned, Tuning=QLoRA2026.06 | 100 | 100 | 100 | 95.2 | 3 | |
| GLAREBackbone=Qwen 2.5 7B, Setting=Fine-tuned, Tuning=QLoRA2026.06 | 100 | 97.6 | 97.6 | 93 | 3 | |
| GLAREBackbone=Qwen 2.5 14B, Setting=Fine-tuned, Tuning=QLoRA2026.06 | 100 | 100 | 100 | 95.4 | 2.6 | |
| GLAREBackbone=Qwen 2.5 0.5B, Setting=Fine-tuned, Tuning=QLoRA2026.06 | 93.8 | 4.6 | 4.6 | 4.4 | 0 | |
| Qwen 2.5 7BSetting=base2026.06 | 81.6 | 37.8 | 37.8 | 3.2 | 3.4 | |
| Gemma 2 2BSetting=base2026.06 | 71.2 | 6.2 | 6.2 | 0 | 0.4 | |
| Qwen 2.5 14BSetting=base2026.06 | 66.2 | 15.6 | 15.6 | 4.4 | 2 | |
| Qwen 2.5 0.5BSetting=base2026.06 | 41.4 | 2.2 | 2.2 | 0 | 0.4 |