Support maze dataset tokenizers update

understanding-search / maze-transformer

This repo is built to facilitate the training and analysis of autoregressive transformers on maze-solving tasks.

24 stars 6 forks source link

Support maze dataset tokenizers update #214

Open aaron-sandoval opened 4 months ago

aaron-sandoval commented 4 months ago

Update to support maze-dataset #38

https://github.com/understanding-search/maze-dataset/pull/38
checks.yml:
all-tok-checks.yml:

review-notebook-app[bot] commented 4 months ago

Check out this pull request on

See visual diffs & provide feedback on Jupyter Notebooks.

Powered by ReviewNB

aaron-sandoval commented 2 months ago

@mivanit All but 2 unit tests are fixed. The last 2 are both zanj tests comparing model outputs before/after a round trip save/load. I'm a bit at a loss on these, I think they could use your eyes. Since 2/4 of the zanj tests pass, it seems more likely due to some other update rather than anything to do with tokenizers.

mivanit commented 1 month ago

so integration tests are hanging after they are complete.

pretty sure I've narrowed this down to a wandb issue, since 8311eb8 makes test_eval_model.py not hang just by moving test_model_loading