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 the issue with media_offset in owl3 when batch_size > 1. #2100

Merged
merged 2 commits into from
Sep 23, 2024

Conversation

LukeForeverYoung
Copy link
Contributor

PR type

  • Bug Fix
  • New Feature
  • Document Updates
  • More Models or Datasets Support

PR information

Write the detail information belongs to this PR.

When the batch size > 1, the media_offset in mPLUG-Owl3 needs to be padded. We use the last value of each media_offset to fill the padding. Additionally, note that the images in the same batch are stacked in order, so the value of media_offset should be incremented by the number of images preceding it

@Jintao-Huang
Copy link
Collaborator

Thank you very much.

@yingdachen yingdachen merged commit 9cc72e3 into modelscope:main Sep 23, 2024
1 of 2 checks passed
Jintao-Huang pushed a commit that referenced this pull request Sep 23, 2024
* mplugowl3 mediaoffset issue

* padding of mediaoffset
tastelikefeet added a commit to tastelikefeet/swift that referenced this pull request Sep 26, 2024
* commit '57b3b9e46aa01bdc5c29b5e3d1e2da0582c9b282': (23 commits)
  fix not impl bug (modelscope#2134)
  Support fine-tuning MLLama. (modelscope#2132)
  Support for fine-tuning and deployment of the Llama 3.2 series models. (modelscope#2130)
  support got-ocr2 (modelscope#2123)
  [TorchAcc] fix: fix find_labels and can_return_loss (modelscope#2120)
  fix qwen2-audio (modelscope#2116)
  Fix qwen2-vl zero2/3 (modelscope#2114)
  support vllm & qwen2-vl video (modelscope#2110)
  Support for fine-tuning Llama 3.1 Omni. (modelscope#2106)
  fix infer device_map (modelscope#2105)
  fix cpu infer device_map (modelscope#2103)
  fix dataset preprocess (modelscope#2102)
  fix deploy openai compat (modelscope#2101)
  Fix the issue with media_offset in owl3 when batch_size > 1. (modelscope#2100)
  fix vllm tokenizer (modelscope#2099)
  Support for fine-tuning Pixtral-12B. (modelscope#2090)
  fix multiprocess remove_columns (modelscope#2088)
  fix qwen2.5 template (modelscope#2081)
  dynamic vit gradient_checkpointing (modelscope#2071)
  Support Mistral-small-inst-2409 (modelscope#2077)
  ...
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.

3 participants