Open
Description
🌟 Feature Description
Currently, the RD-Agent workflow relies heavily on JSON mode to simplify implementation.
However, many LLMs do not natively support the JSON model.
RD-Agent should offer a fallback solution, even though it might result in a higher error rate.
A very naive proposal
- The json mode should be transparent and make the interface consistant with current design.
- It may internelly add some prompts like "you should return only in json format as described in previous content."
- Do necessary post processing like strip ```` fence