We’ve been using scispacy model 'en_core_ci_lg' dfor about 4 years now, but it does not install under Python 3.13.
While this issue is transient and likely to be resolved eventually, it has made us reconsider betting on this model as our base. We’re now unable to retrain our model, if needed, which will eventually become necessary as we are required to move to new Python releases.
Do you know of other models we can use in place of scispacy ‘en_core_ci_lg'?
Hi! Are you able to use a virtual environment with a lower version of Python in the meantime? That shouldn't have any downsides compared to Python 3.13 and will let you keep working. In general, it's often recommended for development to wait before adopting the latest Python version so the ecosystem can catch up and you don't run into incompatibilities (especially since many packages are community-maintained).
If you're building on top of the scispaCy model and are only using part of its predictions, and have a lot of raw data another option could be to use it to build your own data assets so you can train from scratch and aren't dependent on any particular base model. So basically, you'd use the model to label data for you automatically and then use that output to train a model from scratch. You can optionally review samples of it in Prodigy to make sure the predictions are good, and correct them if necessary. Depending on your use case, you might even be able to boost accuracy this way.