big problem: not support token <
I find this model can not be parse to less than symbol (<)
like: question: get people name with age less 25 table: id, name, age
output: SELECT name FROM table WHERE age 25
@JacksonLark
Thank you very much for the report. I will check what happened in this case.
As you suspect, the base model tokenizer also treat < token as a . Maybe I need to try with other model for text to SQL tasks. Thank you very much sharing.
I use Bart-Larger model according to your data to train, hf-model: https://huggingface.co/LarkAI/bart_large_nl2sql
input: question: get people name with age less 25 table: id, name, age
output: SELECT name FROM table WHERE age < 25
I see. Thank you so much for sharing!
Somehow, > symbol is working just fine. I even tried using words "over" "above" and "more than" for this and found correct results.
input: 'get people name with age 28 years with salary over 200000' table: id, name, age, company, salary
output: SELECT name FROM table WHERE age = 28 years AND salary > 200000
tokenizer.decode(token_ids=outputs[0]).replace('<unk>', '<').replace('<pad>', '').replace('</s>', '').strip()
Hi,
Will this work for multiple tables, if so how can I structure the input question string to get JOIN conditions in the output string across the multiple tables ?(complete noob in DL,ML)
Hi,
Will this work for multiple tables, if so how can I structure the input question string to get JOIN conditions in the output string across the multiple tables ?(complete noob in DL,ML)
Hi @moon-cake22 , you can check this one https://huggingface.co/juierror/flan-t5-text2sql-with-schema-v2. It support multiple table and "<" sign.