Added ability to load local models, added early stopping, remove vocab check, fixed GPTJ model conversion #38
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
add pip install
…op early, added the ability to not wait for process.
…gs during finetuning
merge main into branch
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Outdated suggestions cannot be applied.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.
This PR is built on top of the pip package PR. Merge that PR first and if that PR changes those changes will need to be merged first.
In this PR I added the ability to load a model from a local path.
I also added early stopping allowing one to stop based on some token. Related to this, I changed the default behavior of waiting for the subprocess. Using these two together, I was able to achieve a large speedup for my desired task.
I removed the check for vocab size. When finetuning the vocab size can change, especially for GPTJ. Figuring out this was causing issues was a source of headache and is needed for fientuned models.
I also made it so that instead of using the Huggingface vocab and config, if we give a local path we will use those files in that folder, again common for finetuning.
This repo was the only one I could get working for GPTJ. Other repos are using a different GGML format. However, those other repos keep the model in memory with pybindings, such as pyllamacpp. If that feature was added, this repo would be great as suggested #36