Boan

A problem with the input parameter retrieved_duplicates of the avg_prec method

1

The input retrieved_duplicates seems to be unordered. The metric of MAP is related to the order of input. Please let me know if I have misunderstood.

Hello. Thank you for your great work. I can't reproduce the result when using the pose2video model. Can you provide any advice to help me resolve this issue? **command** `...

视频分割偏移问题

5

发现用“python -m tools.scene_cut.cut ${ROOT_META}/xxx.csv --save_dir ${ROOT_CLIPS}”，进行视频分割的时候，分割出来的片段存在偏移。而直接使用scenedetect的split_video_ffmpeg方法，则没有问题。 上述问题会导致镜头不完整，或者跨镜头，不知道这个对训练会不会有什么影响。

Does Quantization (FP8) support the Llama3.2-90B-Vision-Instruct model?

1

Hello, I encountered some problems when loading the Llama3.2-90B-Vision-Instruct model with FP8. Can you help me take a look? Version of llama_stack and llama_models: ``` llama_models == 0.0.41 llama_stack ==...