[Question] How to use cached memory during finetuning ?

[astariul]

As mentioned here, cached memory is not used at finetuning time.

It is also mentioned that one can increase the maximum length at finetuning time, since relative position embeddings are used. However, increasing the size make the model slower (right ?).

---

How can one changes the existing examples (run_squad) to make the model use cached memory at finetuning time ?

[astariul]

@zihangdai Any update on this issue ? Do you plan to release an example for such a use, or is it still not planned ?