| includes: projects/task/crosstask.yaml | |
| model: | |
| model_cls: MMFusionSeparateActionLocalization | |
| mm_encoder_cls: | |
| video_encoder_cls: MMBertForEncoder | |
| text_encoder_cls: BertModel # dummy, not used. | |
| num_hidden_video_layers: 6 | |
| fairseq: | |
| checkpoint: | |
| restore_file: runs/task/checkpoint_best.pt # overwrite the default of VLM. | |