Image Captioning on COCO (COCO Metric)
15.8COCO ScoreKVCapsule
Evaluation Results
| Method | Links | |
|---|---|---|
| KVCapsuleVLM Backbone=Qwen3-VL-8B-Instruct2026.05 | 15.8 | |
| Baseline: all KVVLM Backbone=Qwen3-VL-8B-Instruct2026.05 | 15.7 | |
| PrumergeVLM Backbone=Qwen3-VL-8B-Instruct2026.05 | 15.7 | |
| SparseVLMsVLM Backbone=Qwen3-VL-8B-Instruct2026.05 | 15.6 | |
| SparseVLMsVLM Backbone=llava-v1.6-mistral-7b-hf2026.05 | 15.6 | |
| Baseline: all KVVLM Backbone=Qwen2.5-VL-7B-Instruct2026.05 | 15.4 | |
| SparseVLMsVLM Backbone=Qwen2.5-VL-7B-Instruct2026.05 | 15.4 | |
| PrumergeVLM Backbone=Qwen2.5-VL-7B-Instruct2026.05 | 15.4 | |
| FastVVLM Backbone=llava-v1.6-mistral-7b-hf2026.05 | 15.24 | |
| FastVVLM Backbone=Qwen2-VL-7B-Instruct2026.05 | 15 | |
| Baseline: all KVVLM Backbone=llama3-llava-next-8b-hf2026.05 | 14.7 | |
| KVCapsuleVLM Backbone=llama3-llava-next-8b-hf2026.05 | 14.7 | |
| FastVVLM Backbone=llama3-llava-next-8b-hf2026.05 | 14.5 | |
| KVCapsuleVLM Backbone=llava-v1.6-mistral-7b-hf2026.05 | 14.42 | |
| KVCapsuleVLM Backbone=Qwen2-VL-7B-Instruct2026.05 | 14.4 | |
| KVCapsuleVLM Backbone=Qwen2.5-VL-7B-Instruct2026.05 | 14.4 | |
| SparseVLMsVLM Backbone=llama3-llava-next-8b-hf2026.05 | 14.4 | |
| Baseline: all KVVLM Backbone=llava-v1.6-mistral-7b-hf2026.05 | 14.39 | |
| PrumergeVLM Backbone=llava-v1.6-mistral-7b-hf2026.05 | 14.3 | |
| PrumergeVLM Backbone=llama3-llava-next-8b-hf2026.05 | 14.3 | |
| Baseline: all KVVLM Backbone=Qwen2-VL-7B-Instruct2026.05 | 14.28 | |
| SparseVLMsVLM Backbone=Qwen2-VL-7B-Instruct2026.05 | 14.28 | |
| PrumergeVLM Backbone=Qwen2-VL-7B-Instruct2026.05 | 14.28 | |
| FastVVLM Backbone=Qwen3-VL-8B-Instruct2026.05 | 8.01 | |
| FastVVLM Backbone=Qwen2.5-VL-7B-Instruct2026.05 | 5.9 |