Performance is too low compared to InternVideo2-Stage2_1B-224p-f4
#2
by
kylemin
- opened
I tested the models on the same dataset, but the performance of the 6B model is so bad compared to the 1B model. I used the same code: https://github.com/OpenGVLab/InternVideo/tree/main/InternVideo2/multi_modality
Did you change the code or config when training the 1B and 6B models?
I checked this code (https://huggingface.co/OpenGVLab/InternVideo2-Stage2_6B/blob/main/modeling_internvideo2.py), but could not find any big differences.
I have fixed it, try again, or use the origin ckpt: https://huggingface.co/OpenGVLab/InternVideo2-Stage2_6B-224p-f4
Thanks so much!
kylemin
changed discussion status to
closed