Add example YAML file for training Mistral using DPO #2029
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
NanoCode012
requested changes
Nov 11, 2024
Co-authored-by: NanoCode012 <kevinvong@rocketmail.com>
Adding qlora and removing role-related data (unecessary)
|
I think I addressed all your concerns @NanoCode012. Let me know if there is anything else :) |
NanoCode012
reviewed
Nov 12, 2024
Co-authored-by: NanoCode012 <kevinvong@rocketmail.com>
|
Cool, I commited your comments @NanoCode012 |
NanoCode012
approved these changes
Nov 12, 2024
|
Thank you very much for the PR! |
|
You're welcome! Glad to help. |
bursteratom
pushed a commit
that referenced
this pull request
Nov 18, 2024
* Add example YAML file for training Mistral using DPO * chore: lint * Apply suggestions from code review Co-authored-by: NanoCode012 <kevinvong@rocketmail.com> * Update mistral-dpo.yml Adding qlora and removing role-related data (unecessary) * Rename mistral-dpo.yml to mistral-dpo-qlora.yml * Apply suggestions from code review Co-authored-by: NanoCode012 <kevinvong@rocketmail.com> --------- Co-authored-by: Wing Lian <wing.lian@gmail.com> Co-authored-by: NanoCode012 <kevinvong@rocketmail.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
I provided a YAML file that can be used as an example to perform DPO on a mistral model, using open source datasets and open source models.
Motivation and Context
This is required as the chat template used originally in a Mistral 7B model does not allow for us to regularly perform DPO on the model. The original chat template enforces iterating between user and assistant. However in the chosen and rejected columns of DPO, we only have an assistant turn. For reference, the original chat template can be found in this file. I gave an example where I use the chatml template instead and introduced the respective special tokens, which allows for training.
How has this been tested?
I ran the yaml file locally and trained the model. I merged the model and then tested if the model produced outputs and was successfully using the new special tokens. The whole testing environment is a regular conda environment with axolotl and its dependencies. My changes do not affect other areas of the code since it is a yaml file.
Screenshots (if appropriate)
Types of changes
None
Social Handles (Optional)