NeMo icon indicating copy to clipboard operation
NeMo copied to clipboard

Minor fix in gemma 2 ckpt conversion to support Gemma 2B

Open shashank3959 opened this issue 4 months ago • 0 comments

What does this PR do ?

Add a one line overview of what this PR aims to accomplish.

Collection: [Note which collection this PR will affect]

Changelog

  • Change uses "sliding_window" arg in HF ckpt config to inform window_size. This makes it compatible with all sizes: 2B, 9B and 27B
  • Gemma 2 9B, and Gemma 2 27B HF ckpts have "sliding_window" and "sliding_window_size" both set to 4096
  • Gemma 2 2B config.json in HF ckpt does not have a "sliding_window_size" arg, and only has "sliding_window"

Usage

  • You can potentially add a usage example below
python3 /opt/NeMo/scripts/checkpoint_converters/convert_gemma_hf_to_nemo.py \
--input_name_or_path /path/to/gemma2/checkpoints/hf/2b \
--output_path /path/to/gemma2-2b.nemo \
--tokenizer_path /path/to/tokenizer.model \
[--cpu]

GitHub Actions CI

The Jenkins CI system has been replaced by GitHub Actions self-hosted runners.

The GitHub Actions CI will run automatically when the "Run CICD" label is added to the PR. To re-run CI remove and add the label again. To run CI on an untrusted fork, a NeMo user with write access must first click "Approve and run".

Before your PR is "Ready for review"

Pre checks:

  • [x] Make sure you read and followed Contributor guidelines
  • [ ] Did you write any new necessary tests?
  • [ ] Did you add or update any necessary documentation?
  • [ ] Does the PR affect components that are optional to install? (Ex: Numba, Pynini, Apex etc)
    • [ ] Reviewer: Does the PR have correct import guards for all optional libraries?

PR Type:

  • [ ] New Feature
  • [x] Bugfix
  • [ ] Documentation

If you haven't finished some of the above items you can still open "Draft" PR.

Who can review?

Anyone in the community is free to review the PR once the checks have passed. Contributor guidelines contains specific people who can review PRs to various areas.

Additional Information

  • Related to # (issue)

shashank3959 avatar Oct 19 '24 09:10 shashank3959