LMM Inference Efficiency on 8x RTX 4090 GPUs
1.93FLOPs (T)LLaVA-FA-2B
Evaluation results
| Method | Links | ||
|---|---|---|---|
| LLaVA-FA-2B#I (Number of image tokens)=5762026.01 | 1.93 | 30.6 | |
| DeepSeek-VL-1.3B#I (Number of image tokens)=5762026.01 | 2.01 | 35.1 | |
| Imp-2B#I (Number of image tokens)=5762026.01 | 2.14 | 36.2 | |
| MoE-LLaVA-2B#I (Number of image tokens)=5762026.01 | 2.48 | 39.3 | |
| Bunny-2B#I (Number of image tokens)=5762026.01 | 2.81 | 40.1 | |
| Mini-Gemini-2B#I (Number of image tokens)=5762026.01 | 2.95 | 41.5 |