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

[FIX] Fix multi-modal training #648

Merged
merged 5 commits into from
Sep 21, 2023
Merged

Conversation

lianqing11
Copy link
Collaborator

@lianqing11 lianqing11 commented Sep 21, 2023

  1. support multi-gpu training for multi-modal training
  2. Support Deepspeed's "zero3" training mode for the model in hf_encoder_decoder.py

Copy link
Contributor

@research4pan research4pan left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Added features:

  • Deepspeed zero3 support
  • Add alignment for data type/weight type, converting them automatically to match each other (data type will be converted to the same types as weights)
  • Add alignment for data device/weight device, data devices will be converted to the same device as weights. This enables multi-gpu features.
  • Add an improved dataset data/llava_instruct_80k_truncated.json, which removes very-long-sentence samples from the original dataset.

Copy link
Contributor

@research4pan research4pan left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM, thanks!

@research4pan research4pan merged commit c6b2f14 into main Sep 21, 2023
2 checks passed
@research4pan research4pan deleted the lianqing/multi_modal_training branch March 31, 2024 09:41
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