Closed qyc-98 closed 1 month ago
Hi, thanks for your wonderful work! I'd like to ask how to implement the Gumbel softmax in Tselector for T-selector's training.( in this code)
My implementations are:
Most are correct, but "hard" should be set as False, as indicated here
We simply replace the softmax operation with: video_token_scores = F.gumbel_softmax(video_token_logits, tau=1, hard=False)
video_token_scores = F.gumbel_softmax(video_token_logits, tau=1, hard=False)
Hi, thanks for your wonderful work! I'd like to ask how to implement the Gumbel softmax in Tselector for T-selector's training.( in this code)
My implementations are: