Science Question Answering on SQA (SQA metric)
96.6SQA ScoreLLaVA-OneVision 7B
Evaluation Results
| Method | Links | |
|---|---|---|
| LLaVA-OneVision 7B# Vis tok.=24002025.04 | 96.6 | |
| FLARE-X 8B# Vis tok.=14002025.04 | 91.2 | |
| FLARE-L 8B# Vis tok.=6302025.04 | 89.7 | |
| MiniCPM-Llama3-V-2.5 8B# Vis tok.=4002025.04 | 89.2 | |
| Phi 3.5-Vision# Vis tok.=21002025.04 | 89 | |
| Qwen2.5VL 7B# Vis tok.=14002025.04 | 89 | |
| FLARE-X 3B# Vis tok.=14002025.04 | 87.6 | |
| FLARE-L 3B# Vis tok.=6302025.04 | 86.6 | |
| Florence-VL 8B# Vis tok.=5762025.04 | 85.9 | |
| GRACEData %=20%2026.05 | 85 | |
| Florence-VL 3B# Vis tok.=5762025.04 | 84.6 | |
| RandomData %=20%2026.05 | 84.5 | |
| CADCData %=20%2026.05 | 84.4 | |
| Eagle 8B# Vis tok.=10242025.04 | 84.3 | |
| StepmaxData %=20%2026.05 | 83.8 | |
| LESSData %=20%2026.05 | 83.1 | |
| LongestData %=20%2026.05 | 83 | |
| GRACEData %=5%2026.05 | 83 | |
| ICONSData %=20%2026.05 | 82.6 | |
| GRACEData %=15%2026.05 | 82.6 | |
| GRACEData %=10%2026.05 | 82.5 | |
| Qwen2.5VL 3B# Vis tok.=14002025.04 | 81.4 | |
| MiniCPM-V-2.0 3B# Vis tok.=4002025.04 | 80.7 | |
| Cambrain 8B# Vis tok.=5762025.04 | 80.4 | |
| FullData %=100%2026.05 | 80.2 | |
| Molmo 7B# Vis tok.=12002025.04 | 74.6 | |
| EOSRetain (visual tokens)=322026.06 | 63.83 | |
| SparseVLM-v2Retain (visual tokens)=322026.06 | 63.21 | |
| SparseVLM-v1Retain (visual tokens)=322026.06 | 62.32 | |
| EOSRetain (visual tokens)=162026.06 | 61.35 | |
| ToMeRetain (visual tokens)=322026.06 | 60.8 | |
| SparseVLM-v2Retain (visual tokens)=162026.06 | 60.74 | |
| SparseVLM-v1Retain (visual tokens)=162026.06 | 60.35 | |
| ToMeRetain (visual tokens)=162026.06 | 58.42 | |
| PoolRetain (visual tokens)=322026.06 | 56.1 | |
| PoolRetain (visual tokens)=162026.06 | 53.21 |