Currently, remember returns up to limit items from memory. This leads to suboptimal results if the memory items are short. As an example, MicroGPT will forget some previous commands after 5 steps even though the context could easily fit 10 items.
In an ideal implementation (for an autonomous agent), ThinkGPT would return the maximum possible amount of memory items such that the context window is fully utilized (>= some max_tokens).
Currently,
rememberreturns up tolimititems from memory. This leads to suboptimal results if the memory items are short. As an example, MicroGPT will forget some previous commands after 5 steps even though the context could easily fit 10 items.In an ideal implementation (for an autonomous agent), ThinkGPT would return the maximum possible amount of memory items such that the context window is fully utilized (>= some max_tokens).