Comprehensive guide or course to model finetuning/training results

Hi, it seems there's a lot of great advice hiding in individual threads, but I just wish the information was a little more consolidated -- perhaps a guide of some sort that can give me a rough idea of what effect changing a parameter has on the performance of a model.

Additionally, is there any place to find info on the notation of the model training information?


Hi @joe!

Have you searched or used the spaCy GitHub discussion site yet?

Prodigy offers some training functions (e.g., prodigy train) but these are mainly simple defaults for spacy train. This website is more dedicated to Prodigy-related support (e.g., recipes, Prodigy bugs, etc.). The spaCy GitHub discussion site focuses more on training and may be more helpful for your questions. For instance, here are examples of posts from the spaCy discussion site:

  1. FAQ: Guide to understanding hyperparameters in spaCy
  2. Fine tuning Vs. Training Spacy NER
  3. More information on transformers and hyperparameter tuning

Be sure to also check out that forum's FAQ / Common Issues / Troubleshooting Guide first. It has great suggestions on writing good questions that will lead to faster assistance from the spacy core team. The guide also has some additional helpful links on training too.

In addition to that forum, if you want to dig deeper into training, you likely need to customize your spacy config files. This may be a bit of a learning curve but well worth the time if you want to dig into more advanced training techniques.

I hope this helps!

Great answer, thank you for your reply, these links are very helpful. I quickly realise this question is off topic for this forum. sorry for cluttering. Have a good day.