Feature Request: Size-based composition #3810
Description
📝 Overview
Currently, reflection occurs on an interval basis. LLM has a limit of max token. There is a high possibility that max token will occur on an interval basis. If you compare before max token occurs based on size, you will be able to lower the problem of llm bursting due to max token.
🎯 Proposed Feature
Description: For example, if possible, calculate the current input to detect that it is adjacent to the max token and perform reflection.
🛠️ Benefits
The probability of the problem due to the maximum input token may be further reduced.