Recipe to add a new language to Llama2 #429
Conversation
|
Hi @rahul-sarvam! Thank you for your pull request and welcome to our community. Action RequiredIn order to merge any pull request (code, docs, etc.), we require contributors to sign our Contributor License Agreement, and we don't seem to have one on file for you. ProcessIn order for us to review and merge your suggested changes, please sign at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need to sign the corporate CLA. Once the CLA is signed, our tooling will perform checks and validations. Afterwards, the pull request will be tagged with If you have received this in error or have any questions, please contact us at cla@meta.com. Thanks! |
|
Thank you for signing our Contributor License Agreement. We can now accept your code for this (and any) Meta Open Source project. Thanks! |
There was a problem hiding this comment.
Thanks a lot @rahul-sarvam , it looks great! just added some minor comments.
recipes/multilingual/README.md
Outdated
|
|
||
| ## Continual pre-training | ||
| OpenHathi uses a two-stage pre-training process: | ||
| - Phase 1: learn to translate paragraphs of text (use translated text as context and generate the original text) |
There was a problem hiding this comment.
can we please add how much data we would need/ produce in phase 1 and 2?
There was a problem hiding this comment.
Added in 09028bf
recipes/multilingual/README.md
Outdated
|
|
||
| ``` | ||
| from datasets import load_dataset | ||
| ds = load_dataset("rahular/varta", split="validation", streaming=True) |
There was a problem hiding this comment.
wonder if this would be validation set or train? in terms of supporting enough data for training.
Can we please also have some hint how to save this data in what format to use later, would it be HF dataset etc.?
There was a problem hiding this comment.
Changed in 09028bf
| ``` | ||
|
|
||
| ### Phase 2 | ||
| This is almost the same as phase 1, except that we have to replace the original sentences in an alternating manner to get the data in the required format. We can use the `split_sentences` ([link](https://github.com/AI4Bharat/IndicTrans2/blob/main/huggingface_interface/example.py#L60])) and `batch_translate` ([link](https://github.com/AI4Bharat/IndicTrans2/blob/main/huggingface_interface/example.py#L109)) functions to do this. |
There was a problem hiding this comment.
similarly here, can we also have hint/ add code how to save the dataset/ format.
There was a problem hiding this comment.
Added in 09028bf
Co-authored-by: Hamid Shojanazeri <hamid.nazeri2010@gmail.com>
Co-authored-by: Hamid Shojanazeri <hamid.nazeri2010@gmail.com>
Co-authored-by: Hamid Shojanazeri <hamid.nazeri2010@gmail.com>
What does this PR do?
This PR introduces a new recipe to augment a new language to the Llama2 tokenizer and continue pre-training. This recipe replicates the results of OpenHathi when the right data mixture is used.
Before submitting
Pull Request section?
to it if that's the case.