Layout prediction on HRS numerical
93.28PrecisionQwen3-8B + DivCon
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Qwen3-8B + DivConModel Size=8B, Precision=INT42024.03 | 93.28 | 87.76 | 90.44 | 74.77 | |
| Mistral-7B + DivConModel Size=7B, Precision=INT42024.03 | 93.12 | 87.45 | 90.2 | 69.42 | |
| Qwen3-4B + DivConModel Size=4B, Precision=FP162024.03 | 88.94 | 80.95 | 84.76 | 65.21 | |
| GPT-4 (reference)Model Size=>70B est.2024.03 | 88.13 | 90.91 | 89.5 | 74.05 | |
| Mistral-7BModel Size=7B, Precision=FP162024.03 | 87.47 | 67.73 | 76.35 | 44.86 | |
| TinyLlama + DivConModel Size=1.1B, Precision=INT42024.03 | 81.56 | 65.14 | 72.43 | 31.11 | |
| Phi-2 + DivConModel Size=2.7B, Precision=INT42024.03 | 81.45 | 53.87 | 64.85 | 34.29 | |
| Qwen3-4BModel Size=4B, Precision=FP162024.03 | 76.56 | 67.21 | 71.61 | 45.15 | |
| Phi-2Model Size=2.7B, Precision=FP162024.03 | 76.36 | 51.43 | 61.46 | 23.08 | |
| Qwen3-8BModel Size=8B, Precision=INT42024.03 | 73.27 | 71.16 | 72.2 | 61.16 | |
| TinyLlamaModel Size=1.1B, Precision=FP162024.03 | 71.8 | 43.04 | 53.82 | 17.87 |