Multimodal Video Inference on LLaVA Video Sequence 1.6
2.2Compression Ratio (×)Sali-Cache
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Sali-CacheBackbone=Mistral 7B, Precision=Mixed (FP16, INT8, INT4)2026.02 | 2.2 | — | — | |
| H2OBackbone=Mistral 7B, Precision=FP162026.02 | 1.05 | — | — | |
| BaselineBackbone=Mistral 7B, Precision=FP162026.02 | 1 | — | — |