Multi-modal Benchmark Evaluation on MMBench v1.0 (dev)
69.2MMB^dev ScoreMobileVLM V2 7B
Evaluation results
| Method | Links | |
|---|---|---|
| MobileVLM V2 7BLLM=Vicuna-7B, #Samples=3.6M2026.03 | 69.2 | |
| ShareGPT4VLLM=Vicuna-7B, #Samples=1.9M2026.03 | 68.8 | |
| LLaVA-1.5LLM=Vicuna-7B, #Samples=1.2M2026.03 | 64.3 | |
| Cosine KD w/ Beta-KD (Instance)LLM=MobileLLaMA 1.4B, #Samples=3.6M2026.03 | 60.2 | |
| MoE-LLaVA-2BLLM=Qwen-1.5-1.8B, #Samples=2.2M2026.03 | 59.7 | |
| MobileVLM 3BLLM=MobileLLaMA 2.7B, #Samples=3.6M2026.03 | 59.6 | |
| MoE-LLaVA-1.6BLLM=StableLM-1.6B, #Samples=2.2M2026.03 | 59.4 | |
| Align-KD w/ Beta-KD (Instance)LLM=MobileLLaMA 1.4B, #Samples=3.6M2026.03 | 59.1 | |
| Cosine KD w/ Beta-KD (Task)LLM=MobileLLaMA 1.4B, #Samples=3.6M2026.03 | 59.1 | |
| Align-KD w/ Beta-KD (Task)LLM=MobileLLaMA 1.4B, #Samples=3.6M2026.03 | 58.8 | |
| MobileVLM -V2 1.7BLLM=MobileLLaMA 1.4B, #Samples=3.6M2026.03 | 57.6 | |
| Cosine KDLLM=MobileLLaMA 1.4B, #Samples=3.6M2026.03 | 57.1 | |
| Align-KDLLM=MobileLLaMA 1.4B, #Samples=3.6M2026.03 | 57 | |
| LLAVADILLM=MobileLLaMA 1.4B, #Samples=3.6M2026.03 | 55 | |
| MobileVLM 1.7BLLM=MobileLLaMA 1.4B, #Samples=3.6M2026.03 | 53.2 |