Currently the template used by the all featurizers are in a QA format suitable for SFT. But if the goal is tot translate fuzzy natural-language descriptions into feature vectors, we need a different set of templates. The best case scenario is to have a couple of different templates custom -made for each template.
Currently the template used by the all featurizers are in a QA format suitable for SFT. But if the goal is tot translate fuzzy natural-language descriptions into feature vectors, we need a different set of templates. The best case scenario is to have a couple of different templates custom -made for each template.