Hello all,
I trained the T5-base model using gretelai/synthetic_text_to_sql data set and then fine tuned it on my specific table schema and set of example queries.
When I test the fine-tuned model it keeps missing the “<” token in the generated query results.
I have played with various fine-tuning params – like number of epochs.
Why thus the resultant model not know to use the “<” token ?
I added a couple of SQL examples with explicit “<” to the dataset but when I query back it gives me the resulting SQL without the “<” in it which is an incorrect SQL!
Cheers.
1 Like
Hmm… Perhaps tokenizer vocab issue?
https://stackoverflow.com/questions/75851029/t5-fine-tuned-model-outputs-unk-instead-of-curly-braces-and-other-special-char