Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[ssl/w2vbert] support part of w2vbert training #2039

Merged
merged 5 commits into from
Oct 10, 2023
Merged

Conversation

Mddct
Copy link
Collaborator

@Mddct Mddct commented Oct 9, 2023

ref: https://arxiv.org/pdf/2108.06209v2.pdf
note: bestrq >= w2vbert >= hubert >= wav2vec2 from some public information,

btw: HuBERT, which relies on an iterative re-clustering and re-training process, so no plan to implement hubert for now

Screenshot 2023-10-09 at 17 07 53

@Mddct Mddct changed the title [ssl/wavbert] support part of w2vbert training [ssl/w2vbert] support part of w2vbert training Oct 9, 2023
@Mddct Mddct force-pushed the Mddct-ssl-w2vbert branch from 1b43df1 to 08c6df0 Compare October 9, 2023 07:41
@Mddct Mddct marked this pull request as ready for review October 9, 2023 07:47
@Mddct Mddct marked this pull request as draft October 9, 2023 07:52
@Mddct Mddct marked this pull request as ready for review October 9, 2023 09:09
wenet/ssl/w2vbert/w2vbert_model.py Show resolved Hide resolved
wenet/ssl/w2vbert/w2vbert_model.py Show resolved Hide resolved
wenet/ssl/w2vbert/w2vbert_model.py Outdated Show resolved Hide resolved
xingchensong
xingchensong previously approved these changes Oct 9, 2023
@xingchensong xingchensong merged commit 35acffc into main Oct 10, 2023
@xingchensong xingchensong deleted the Mddct-ssl-w2vbert branch October 10, 2023 03:06
@xingchensong
Copy link
Member

Great, THX!

@xingchensong
Copy link
Member

xingchensong commented Oct 10, 2023

TODO:

  • 初始化方式补齐下ref(可以稳定encoder训练)
  • x的L2补齐下ref
  • multi-softmax用的bias补充下注释

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants