Document Processing Acceleration on olmOCR-Bench long documents filtered > 8192 tokens
4.08AALQwen3-VL-8B
Evaluation Results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Qwen3-VL-8BParameters=9B, Technique=Hierarchical Speculative Decoding2026.02 | 4.08 | 2.4 | 2.42 | 2.51 | 1.64 | 1.6 | 1.11 | 2.65 | 3.75 | 2.19 | |
| Qwen3-VL-2BParameters=2B, Technique=Hierarchical Speculative Decoding2026.02 | 4.89 | 2.97 | 2.98 | 3 | 1.65 | 2.2 | 1.46 | 3.2 | 4.48 | 2.54 |