Image Captioning on COCO-Cap
55.3ROUGE-LTGV-KV
Evaluation Results
| Method | Links | |
|---|---|---|
| TGV-KVModel=LLaVA-1.5-7B, Retention Ratio=50%2026.06 | 55.3 | |
| SnapKVModel=LLaVA-1.5-7B, Retention Ratio=50%2026.06 | 55.2 | |
| VanillaModel=LLaVA-1.5-7B, Retention Ratio=50%2026.06 | 55.1 | |
| VanillaModel=LLaVA-1.5-7B, Retention Ratio=20%2026.06 | 55.1 | |
| VanillaModel=LLaVA-1.5-7B, Retention Ratio=10%2026.06 | 55.1 | |
| TGV-KVModel=LLaVA-1.5-7B, Retention Ratio=20%2026.06 | 55 | |
| SnapKVModel=LLaVA-1.5-7B, Retention Ratio=20%2026.06 | 54.6 | |
| StreamingLLMModel=LLaVA-1.5-7B, Retention Ratio=50%2026.06 | 54.3 | |
| TGV-KVModel=LLaVA-1.5-7B, Retention Ratio=10%2026.06 | 52.6 | |
| StreamingLLMModel=LLaVA-1.5-7B, Retention Ratio=20%2026.06 | 52.2 | |
| StreamingLLMModel=LLaVA-1.5-7B, Retention Ratio=10%2026.06 | 51 | |
| VanillaModel=Qwen3-VL-8B-Instruct, Retention Ratio=50%2026.06 | 42 | |
| VanillaModel=Qwen3-VL-8B-Instruct, Retention Ratio=20%2026.06 | 42 | |
| VanillaModel=Qwen3-VL-8B-Instruct, Retention Ratio=10%2026.06 | 42 | |
| TGV-KVModel=Qwen3-VL-8B-Instruct, Retention Ratio=50%2026.06 | 42 | |
| StreamingLLMModel=Qwen3-VL-8B-Instruct, Retention Ratio=50%2026.06 | 41.7 | |
| PrefixKVModel=Qwen3-VL-8B-Instruct, Retention Ratio=50%2026.06 | 41.7 | |
| TGV-KVModel=Qwen3-VL-8B-Instruct, Retention Ratio=20%2026.06 | 41.6 | |
| H2OModel=Qwen3-VL-8B-Instruct, Retention Ratio=50%2026.06 | 41.4 | |
| PrefixKVModel=Qwen3-VL-8B-Instruct, Retention Ratio=20%2026.06 | 40.6 | |
| SnapKVModel=Qwen3-VL-8B-Instruct, Retention Ratio=50%2026.06 | 40.4 | |
| ElasticCacheModel=Qwen3-VL-8B-Instruct, Retention Ratio=50%2026.06 | 39.9 | |
| StreamingLLMModel=Qwen3-VL-8B-Instruct, Retention Ratio=20%2026.06 | 39.6 | |
| TGV-KVModel=Qwen3-VL-8B-Instruct, Retention Ratio=10%2026.06 | 39.5 | |
| H2OModel=Qwen3-VL-8B-Instruct, Retention Ratio=20%2026.06 | 39.3 | |
| ElasticCacheModel=Qwen3-VL-8B-Instruct, Retention Ratio=20%2026.06 | 38.1 | |
| H2OModel=Qwen3-VL-8B-Instruct, Retention Ratio=10%2026.06 | 37.9 | |
| SnapKVModel=Qwen3-VL-8B-Instruct, Retention Ratio=20%2026.06 | 36.3 | |
| PrefixKVModel=Qwen3-VL-8B-Instruct, Retention Ratio=10%2026.06 | 36.2 | |
| SnapKVModel=Qwen3-VL-8B-Instruct, Retention Ratio=10%2026.06 | 35.7 | |
| StreamingLLMModel=Qwen3-VL-8B-Instruct, Retention Ratio=10%2026.06 | 35.3 | |
| ElasticCacheModel=Qwen3-VL-8B-Instruct, Retention Ratio=10%2026.06 | 34.6 | |
| PrefixKVModel=LLaVA-1.5-7B, Retention Ratio=50%2026.06 | 34 | |
| H2OModel=LLaVA-1.5-7B, Retention Ratio=50%2026.06 | 17.7 | |
| PrefixKVModel=LLaVA-1.5-7B, Retention Ratio=20%2026.06 | 11.1 | |
| H2OModel=LLaVA-1.5-7B, Retention Ratio=20%2026.06 | 9.8 | |
| ElasticCacheModel=LLaVA-1.5-7B, Retention Ratio=50%2026.06 | 9.6 | |
| ElasticCacheModel=LLaVA-1.5-7B, Retention Ratio=20%2026.06 | 8.7 | |
| ElasticCacheModel=LLaVA-1.5-7B, Retention Ratio=10%2026.06 | 8.4 | |
| H2OModel=LLaVA-1.5-7B, Retention Ratio=10%2026.06 | 8 | |
| PrefixKVModel=LLaVA-1.5-7B, Retention Ratio=10%2026.06 | 7.9 | |
| SnapKVModel=LLaVA-1.5-7B, Retention Ratio=10%2026.06 | 0.9 |