Motion Captioning on EventMind 1.0 (test)
52.3GDCEventFlash-7B
Evaluation results
| Method | Links | |
|---|---|---|
| EventFlash-7BParams=7B, LLM Backbone=Qwen-2.5, Max Bins=1,000, Throughput (Token/s)=24.02026.02 | 52.3 | |
| EventFlash-3BParams=3B, LLM Backbone=Qwen-2.5, Max Bins=1,000, Throughput (Token/s)=28.52026.02 | 46.8 | |
| EventFlash-ZeroParams=3B, LLM Backbone=Qwen-2.5, Max Bins=1,000, Throughput (Token/s)=2.32026.02 | 45.3 | |
| VideoChat2-FlashParams=7B, LLM Backbone=Qwen2.5, Max Bins=1,000, Throughput (Token/s)=n/a2026.02 | 36.2 | |
| VideoChat2-FlashParams=2B, LLM Backbone=Qwen2.5, Max Bins=1,000, Throughput (Token/s)=n/a2026.02 | 31.6 | |
| LLaVA-Next-VideoParams=7B, LLM Backbone=Qwen2.5, Max Bins=56, Throughput (Token/s)=n/a2026.02 | 31.2 | |
| Qwen2.5 VLParams=7B, LLM Backbone=Qwen2.5, Max Bins=768, Throughput (Token/s)=n/a2026.02 | 22.1 | |
| Qwen2.5 VLParams=3B, LLM Backbone=Qwen2.5, Max Bins=768, Throughput (Token/s)=n/a2026.02 | 20.6 | |
| InternVL2.5Params=8B, LLM Backbone=InternLM2.5, Max Bins=n/a, Throughput (Token/s)=n/a2026.02 | 19.7 | |
| InternVL2.5Params=4B, LLM Backbone=Qwen2.5, Max Bins=n/a, Throughput (Token/s)=n/a2026.02 | 17.9 |