Multimodal Evaluation on COCO
100.22Average PercentageFastV
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| FastVBackbone=LLaVA-v1.5-13B, Compression Timing=During prefill, FLOPs (Tera)=12.4, Memory (GB)=0.36, TTFT (sec)=0.085, Vision Token Retention=55%2026.05 | 100.22 | 1.16 | |
| VanillaBackbone=LLaVA-v1.5-13B, FLOPs (Tera)=19.4, Memory (GB)=0.55, TTFT (sec)=0.130, Vision Token Retention=55%2026.05 | 100 | 1.16 | |
| VanillaBackbone=LLaVA-v1.5-7B, FLOPs (Tera)=10.2, Memory (GB)=0.35, TTFT (sec)=0.078, Vision Token Retention=55%2026.05 | 100 | 1.1 | |
| LightKVBackbone=LLaVA-v1.5-13B, Compression Timing=During prefill, FLOPs (Tera)=12.6, Memory (GB)=0.37, TTFT (sec)=0.098, Vision Token Retention=55%2026.05 | 99.94 | 1.15 | |
| LightKVBackbone=LLaVA-v1.5-7B, Compression Timing=During prefill, FLOPs (Tera)=6.6, Memory (GB)=0.23, TTFT (sec)=0.065, Vision Token Retention=55%2026.05 | 99.79 | 1.11 | |
| LightKVBackbone=LLaVA-NeXT-7B, Compression Timing=During prefill, FLOPs (Tera)=22.3, Memory (GB)=0.67, TTFT (sec)=0.259, Vision Token Retention=55%2026.05 | 98.85 | 0.98 | |
| LightKVBackbone=LLaVA-NeXT-13B, Compression Timing=During prefill, FLOPs (Tera)=37.3, Memory (GB)=1.05, TTFT (sec)=0.383, Vision Token Retention=55%2026.05 | 98.12 | 0.96 | |
| HiREDBackbone=LLaVA-v1.5-7B, Compression Timing=During prefill, Vision Token Retention=55%2026.05 | 96.45 | 1.03 | |
| ToMe (P)Backbone=LLaVA-v1.5-13B, Compression Timing=During prefill, FLOPs (Tera)=12.6, Memory (GB)=0.37, TTFT (sec)=0.094, Vision Token Retention=55%2026.05 | 93.96 | 1.16 | |
| PiToMeBackbone=LLaVA-v1.5-13B, Compression Timing=During prefill, FLOPs (Tera)=12.6, Memory (GB)=0.37, TTFT (sec)=0.093, Vision Token Retention=55%2026.05 | 93.42 | 1.14 | |
| ToFuBackbone=LLaVA-v1.5-13B, Compression Timing=During prefill, FLOPs (Tera)=12.6, Memory (GB)=0.37, TTFT (sec)=0.094, Vision Token Retention=55%2026.05 | 93.36 | 1.14 | |
| ToMe (C)Backbone=LLaVA-v1.5-13B, Compression Timing=Post prefill, FLOPs (Tera)=19.0, Memory (GB)=0.33, TTFT (sec)=0.141, Vision Token Retention=55%2026.05 | 87.1 | 1 | |
| ImgRandBackbone=LLaVA-v1.5-13B, Compression Timing=Post prefill, FLOPs (Tera)=19.0, Memory (GB)=0.31, TTFT (sec)=0.134, Vision Token Retention=55%2026.05 | 85.09 | 0.95 | |
| RandBackbone=LLaVA-v1.5-13B, Compression Timing=Post prefill, FLOPs (Tera)=19.0, Memory (GB)=0.31, TTFT (sec)=0.134, Vision Token Retention=55%2026.05 | 70.53 | 0.48 | |
| ElasticBackbone=LLaVA-v1.5-13B, Compression Timing=Post prefill, FLOPs (Tera)=19.3, Memory (GB)=0.31, TTFT (sec)=0.598, Vision Token Retention=55%2026.05 | 68.54 | 0.96 | |
| ElasticBackbone=LLaVA-NeXT-13B, Compression Timing=Post prefill, TTFT (sec)=2.302, Vision Token Retention=55%2026.05 | 0 | — |