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.

Sign up for GitHub

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

Jump to bottom

Add Semantic Factorization #363

Open
10 tasks done
gkielian opened this issue Jan 19, 2025 · 1 comment · May be fixed by #359
Open
10 tasks done

Add Semantic Factorization #363

gkielian opened this issue Jan 19, 2025 · 1 comment · May be fixed by #359
Assignees
@gkielian

Comments

@gkielian
Copy link
Collaborator

  • prepare.py
    • interface for doing tokenization with labelling
  • train.py
    • get batch modification for mulitiple Y targets
    • modify model.py before each Y target for different (or same) LM Head
    • should also automatically just work for multitoken prediction
  • model.py
    • if mode is semantic factorization, then take factorization number to group the inputs into a single embedding vector (sum of inputs being default)
    • create number of LM heads (multitoken output)
    • support for either single or multiple embedding tables
@gkielian gkielian self-assigned this Jan 19, 2025
@gkielian gkielian linked a pull request Jan 19, 2025 that will close this issue
Add multicontext training #359
Open
@gkielian
Copy link
Collaborator Author

Complete first draft for implementing controlled per position semantic embeddings as per https://arxiv.org/abs/2405.20671

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees

@gkielian gkielian

Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging a pull request may close this issue.

Add multicontext training gkielian/ReaLLMASIC_nanogpt
1 participant
@gkielian