New issue

Jump to bottom

[meta] Support supervised fine-tuning (SFT) #106

Closed

2 tasks done

tscholak opened this issue Jan 8, 2025 · 2 comments

Labels

enhancement need update

Milestone

Collaborator

tscholak commented Jan 8, 2025 •

edited by sohamparikh

Loading

The following issues need to be resolved to support fine-tuning:

Tasks

Preview Give feedback

Implement cross-example attention masking to isolate examples during SFT: [feat] Optional Attention Mask to Prevent Cross-Document Attention in Sequences #62
Implement loss masking to exclude prompts and user inputs from the LM loss: [feat] Implement Loss Masking to Exclude Predefined Token Spans from LM Loss #109
Options

tscholak added the enhancement label

tscholak added this to the 0.3.0 milestone

jlamypoirier added the need update label

Collaborator

jlamypoirier commented Jan 28, 2025

Can we close this as redundant?

Collaborator Author

tscholak commented Mar 23, 2025

all done

tscholak closed this as completed

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