Allow "weight: 0" in messages to mask them #1703
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
DavidFarago
force-pushed
the
mr-weight-on-main
branch
2 times, most recently
from
ccebf54
to
d9bbf5d
Compare
winglian
requested changes
Jun 12, 2024
Allow in message objects the additional key `weight`, which can be set to 0 (or 1) to cause that message to be masked out (or left unmasked) for training (similar to [1]). This is helpful for training the model to be robust and capable of error recovery upon a bad assistant message. A missing `weight` key defaults to weight 1, to guarantee downward compatibility. Extend `src/axolotl/prompters.py::_build_result` and `src/axolotl/prompt_strategies/sharegpt.py::SimpleShareGPTPromptTokenizingStrategy::get_conversation_thread` to return the turns with weights as additional tuple element. Do this in axolotl directly instead of modifying `fastchat.conversation`'s `Conversation`. Extend `src/axolotl/prompt_tokenizers.py::tokenize_prompt` to mask out tokens when weight is set to 0. Extend `tests/prompt_strategies/test_sharegpt.py` with four test cases that contain messages with `weight` keys. Switch names `test_w_train_on_input` and `test_no_train_on_input`. [1]: https://github.com/mistralai/mistral-finetune
DavidFarago
force-pushed
the
mr-weight-on-main
branch
from
d9bbf5d
to
14428da
Compare
|
you are welcome, @winglian -- thank you for reviewing. Will you merge this PR or give me write access to this repository? |
|
merged. thanks @DavidFarago ! |
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Outdated suggestions cannot be applied.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.
Allow in message objects the additional key
weight, which can be set to 0 (or 1) to cause that message to be masked out (or left unmasked) for training (similar to 1).Description
Extend
src/axolotl/prompters.py::_build_resultto return the turns with weights as additional tuple element. Do this in axolotl directly instead of modifyingfastchat.conversation'sConversation.Extend
src/axolotl/prompt_tokenizers.py::tokenize_promptto mask out tokens when weight is set to 0.Motivation and Context
This is helpful for training the model to be robust and capable of error recovery upon a bad assistant message. A missing
weightkey defaults to weight 1, to guarantee downward compatibility.How has this been tested?
Extend
tests/prompt_strategies/test_sharegpt.pyto contain messages withweightkeys.