StreamSpeech / fairseq /examples /MMPT /projects /task /crosstask_videoclip.yaml
fasdfsa's picture
init
901e06a
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.