Visual Question Answering on VQAv2, VizWiz, GQA, TextVQA, and MME
64.9GQA AccuracyCogVLM-7B
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| CogVLM-7BSize=~7B, LLM=Vicuna-7B, # Samples=1500 M2025.10 | 64.9 | 82.3 | — | 78.2 | 71.8 | — | |
| Qwen2.5-VL-7BSize=~7B, LLM=Qwen2-7B, # Samples=1500 M2025.10 | 63.9 | 82.8 | 61.7 | 73.9 | 83.9 | 73.2 | |
| LLaVA-7B†Size=~7B, LLM=Qwen2.5-7B, # Samples=1.2 M2025.10 | 63.8 | 81.4 | 50.6 | 64.6 | 77.5 | 67.6 | |
| Imp-3BSize=~4B, LLM=Phi2-2.7B, # Samples=1.5 M2025.10 | 63.5 | 81.2 | 54.1 | 59.8 | 72.3 | 66.2 | |
| MoE-LLaVA-3BSize=~4B, LLM=Phi2-2.7B, # Samples=2.6 M2025.10 | 62.6 | 79.9 | 43.7 | 57 | 71.6 | 63 | |
| Bunny-3BSize=~4B, LLM=Phi2-2.7B, # Samples=2.6 M2025.10 | 62.5 | 79.8 | 43.8 | 56.7 | 74.4 | 63.4 | |
| LLaVA-7B†Size=~7B, LLM=Qwen1.5-7B, # Samples=1.2 M2025.10 | 62.3 | 79.8 | 53.1 | 60.1 | 68.6 | 64.8 | |
| LLaVA-KD-2BSize=~2B, LLM=Qwen1.5-1.8B, # Samples=1.2 M2025.10 | 62.3 | 78.8 | 44.8 | 53.4 | 68.9 | 61.6 | |
| CompoDistill-2BSize=~2B, LLM=Qwen1.5-1.8B, # Samples=1.2 M2025.10 | 62.2 | 78.8 | 42.5 | 56.4 | 70.1 | 61.9 | |
| LLaVA-4B† (Teacher)Size=~4B, LLM=Qwen1.5-4B, # Samples=1.2 M2025.10 | 62.1 | 79.1 | 48 | 56.7 | 67.2 | 62.6 | |
| Imp-2BSize=~2B, LLM=Qwen1.5-1.8B, # Samples=1.5 M2025.10 | 61.9 | 79.2 | 39.2 | 54.5 | 65.2 | 60 | |
| MoE-LLaVA-2BSize=~2B, LLM=Qwen1.5-1.8B, # Samples=2.2 M2025.10 | 61.5 | 76.2 | 32.6 | 48 | 64.6 | 56.6 | |
| Deepseek-VL-7BSize=~7B, LLM=DLLM-7B, # Samples=2000 M2025.10 | 61.3 | — | 49.9 | 64.7 | 73.4 | — | |
| MobileVLM-3BSize=~4B, LLM=MobileLLaMA-2.7B, # Samples=1.3 M2025.10 | 61.1 | — | — | 57.5 | 72 | — | |
| Qwen2.5-VL-3BSize=~4B, LLM=Qwen2-3B, # Samples=1.2 M2025.10 | 60.9 | 80.4 | 54.1 | 68.5 | 72.8 | 67.3 | |
| MiniGemini-2BSize=~2B, LLM=Gemma-2B, # Samples=2.7 M2025.10 | 60.7 | — | 41.5 | 56.2 | 67 | — | |
| Bunny-2BSize=~2B, LLM=Qwen1.5-1.8B, # Samples=2.6 M2025.10 | 59.6 | 76.6 | 34.2 | 53.2 | 65 | 57.7 | |
| Deepseek-VL-1.3BSize=~2B, LLM=DLLM-1.3B, # Samples=2000 M2025.10 | 59.3 | — | 36.8 | 58.4 | 65.3 | — | |
| MobileVLM-1.7BSize=~2B, LLM=MobileLLaMA-1.4B, # Samples=1.2 M2025.10 | 59.3 | — | — | 52.1 | 65.1 | — | |
| LLaVA-MoD-2B†Size=~2B, LLM=Qwen1.5-1.8B, # Samples=5.0 M2025.10 | 58.8 | 76.3 | 43 | 52.7 | 63.4 | 58.9 | |
| LLaVADI-2B†Size=~2B, LLM=Qwen1.5-1.8B, # Samples=1.2 M2025.10 | 58.7 | 75.8 | 35.2 | 49.4 | 63.8 | 56.6 | |
| LLaVA-2B (SFT model)Size=~2B, LLM=Qwen1.5-1.8B, # Samples=1.2 M2025.10 | 57.9 | 73.6 | 31.2 | 50.6 | 61.2 | 54.9 | |
| MiniCPM-VSize=~4B, LLM=MiniCPM-2.4B, # Samples=570 M2025.10 | 52.1 | — | 60.2 | 73.2 | 70.5 | — |