metadata
datasets:
- niltheory/ExistenceTypes
language:
- en
Existence Analysis Model
Created for: Compendium Terminum, IP.
Iteration #1: base distilBert used to train the model with a data training of 96.
Iteration #2: Increased the data training from 96 to 296. Which greatly increased the accuracy scores, albeit there is some edge cases.
Iteration #3:
Switched from bert-base-uncased to bert-large-cased-whole-word-masking. Which it has more contextual sensitivity and is slightly more accurate than Iteration #2. It's overall more nuanced and sensitive.