The model struggles with names and assumes a word. Tostitos is often heard wrong, for example.
Either lean on a larger model and suffer at least 2x inference time, or build a dictionary of common typos and run the output through that before presenting to the users,
The model struggles with names and assumes a word. Tostitos is often heard wrong, for example.
Either lean on a larger model and suffer at least 2x inference time, or build a dictionary of common typos and run the output through that before presenting to the users,