First of all, thank you for all the work on Megumin Suite. The Memory Core provides much better story continuity than other memory extensions I've tried (such as Qvink), so I'd really like to keep using it.
However, I've noticed a significant performance issue on long-running RP chats.
Environment
- Megumin Suite V8
- SillyTavern running on a VPS
- Chat length: ~650 messages
- Browser: Chromium-based (Edge/Chrome)
What happens
Once the chat reaches around 650+ messages, pressing Send introduces a long delay before the generation request is even sent.
Interestingly:
- VPS CPU stays almost idle.
- SillyTavern server CPU also remains very low.
- My proxy does not receive the generation request immediately.
- The browser repeatedly issues
/count requests for token counting before the actual generation starts.
From the browser console I also see: Core/all messages: 643/643
Testing
I tried disabling only Memory Core while leaving the rest of Megumin enabled (NPC Bank, V8 preset, etc.).
Result:
- ✅ Sending messages becomes almost instantaneous.
- ✅ The proxy receives the generation request immediately.
- ✅ UI becomes much more responsive.
I also tested Qvink Memory as you suggested (on reddit). It performs much better, but I personally find Memory Core's story continuity noticeably superior.
Suggestion
If possible, it may be worth optimizing how Memory Core handles long conversations.
For example, reducing the amount of work performed before generation or using a more incremental processing strategy could help preserve its excellent continuity while improving responsiveness on very long chats.
I'd be happy to help test future builds if needed.
First of all, thank you for all the work on Megumin Suite. The Memory Core provides much better story continuity than other memory extensions I've tried (such as Qvink), so I'd really like to keep using it.
However, I've noticed a significant performance issue on long-running RP chats.
Environment
What happens
Once the chat reaches around 650+ messages, pressing Send introduces a long delay before the generation request is even sent.
Interestingly:
/countrequests for token counting before the actual generation starts.From the browser console I also see: Core/all messages: 643/643
Testing
I tried disabling only Memory Core while leaving the rest of Megumin enabled (NPC Bank, V8 preset, etc.).
Result:
I also tested Qvink Memory as you suggested (on reddit). It performs much better, but I personally find Memory Core's story continuity noticeably superior.
Suggestion
If possible, it may be worth optimizing how Memory Core handles long conversations.
For example, reducing the amount of work performed before generation or using a more incremental processing strategy could help preserve its excellent continuity while improving responsiveness on very long chats.
I'd be happy to help test future builds if needed.