Is it correct to keep using adapter_kv_cache during training in litgpt/adapter.py?

[fleetfootwork]

Is it correct to keep using adapter_kv_cache during training in litgpt/adapter.py? I think self.adapter_wte and self.attn are updated during training, so ak and av should not use kv_cache. Thank you very much!
But it seems that during training, self.adapter_kv_cache was also used.

[fleetfootwork]

@Andrei-Aksionov @carmocca Can you answer my doubts? I would greatly appreciate it

[Andrei-Aksionov]

Hello @RookieXwc
Thanks for the question.

To me it also looks strange.
But I would like to have a confirmation from @carmocca that we don't miss anything.

[carmocca]

You are right. It's also missing the clearing of the adapter kv cache. It could be done by overriding https://github.com/Lightning-AI/litgpt/blob/main/litgpt/model.py#L133-L136 for the adapter model

A PR with the fixes would be welcome

[Borda]

likely will be resolved by #2061