You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
it make sense a XML is closely related to HTML which is the dominant form in the training datasets.
This might particularly help weaker LLM.
switching to XML for the model output will improve the performance of the agents (this can be then converted to JSON for better human readability and python usage.
The text was updated successfully, but these errors were encountered:
The current LLM models generate better quality results when tasked to generate XML than JSON see: https://medium.com/@isaiahdupree33/optimal-prompt-formats-for-llms-xml-vs-markdown-performance-insights-cef650b856db
it make sense a XML is closely related to HTML which is the dominant form in the training datasets.
This might particularly help weaker LLM.
switching to XML for the model output will improve the performance of the agents (this can be then converted to JSON for better human readability and python usage.
The text was updated successfully, but these errors were encountered: