Fix mixtral_nemo_to_hf conversion

[tdene]

What does this PR do ?

Fix convert_mixtral_nemo_to_hf.py

Collection: scripts/checkpoint_converters

Changelog

• Fix two typos
• Allow for custom tokenizer to be passed in

PR Type:

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

Who can review?

@cuichenx

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.

[janekl]

Is it fine & really required to introduce extra param --input_tokenizer?

If yes, should we update relevant docs? https://github.com/NVIDIA/NeMo/blob/main/docs/source/ckpt_converters/dev_guide.rst

CC @yaoyu-33 - please have a look, maybe you will have some comments

[tdene]

Is it fine & really required to introduce extra param --input_tokenizer?

I'm modeling this off #8000. In both cases, the problem is that the converter tries to use the default HuggingFace tokenizer, while the custom model that's being converted may well be using a different tokenizer.

Perhaps the correct solution in both cases is to unpack the tokenizer out of the .nemo file during the conversion script, instead of adding a --input_tokenizer argument.

[github-actions[bot]]

This PR is stale because it has been open for 14 days with no activity. Remove stale label or comment or update or this will be closed in 7 days.

[github-actions[bot]]

This PR was closed because it has been inactive for 7 days since being marked as stale.