Hi!
Currently when dev_mode=True, an excpetion is thrown with an error of unexpected token. However, it would be really usefull to be able to get the partial code as well. Then we could use this information in self-refinements loops to generate even better code.
Impact: This is escpecially usefull in cases of very low-resource languages, that LLMs struggles with. Currently for my cases, the LLM gets stuck on producing reserved keywords in unexpected cases (where identifier is excpected)..
Hi!
Currently when dev_mode=True, an excpetion is thrown with an error of unexpected token. However, it would be really usefull to be able to get the partial code as well. Then we could use this information in self-refinements loops to generate even better code.
Impact: This is escpecially usefull in cases of very low-resource languages, that LLMs struggles with. Currently for my cases, the LLM gets stuck on producing reserved keywords in unexpected cases (where identifier is excpected)..