SamChen567/VideoBenchEval-results
phase3_qbase_full_hybrid_lr1e-5：評測與 ablation 結果 截至 2026-10-23（UTC+8）。模型：SamChen567/phase3_qbase_full_hybrid_lr1e-5。評測程式快照在 code/（VideoBenchEval：vbe/infer.py 主評測、vbe/ablate.py ablation）。 三個 benchmark 全集：Video-MME（2,700 題，無字幕）、MLVU dev（2,174 題）、LongVideoBench val（1,337 題）。每支影片的 LLM 端 vision token 預算 B = 1k / 2k / 4k / 8k / 16k。官方 VideoLLaMA3 prompt 與答案解析，greedy、16 個新 token。所有 run 皆完整，無錯誤。 結論 qbase 的價值在高壓縮率。 切段和預算都不變，只把每個壓縮段換成「選一幀再 resize / merge」：1k、2k 時 Video-MME… See the full description on the dataset page: https://huggingface.co/datasets/SamChen567/VideoBenchEval-results.
| Author / organization | SamChen567 |
|---|---|
| License (as declared) | other |
| Downloads | 323 (as reported by Hugging Face) |
| Likes | 1 |
| Created | 22 October 2026 |
| Last updated at source | 23 October 2026 |
More from SamChen567
Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 23 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.