Add memory state management for LLMs

[VictorOdede]

trafficstars

Extend context window using disc storage\RAM\cache\other

[grepfruit19]

I've been playing around with vector databases the past few days. Could take this ticket on if we have more context (assuming that vector databases would be a good case for this?)

[VictorOdede]

I've been playing around with vector databases the past few days. Could take this ticket on if we have more context (assuming that vector databases would be a good case for this?)

Sounds like a great idea! You should join the triage on Monday to discuss this further before getting started

[mmirman]

How do we measure success here?

[julietkeynai]

Related to #304

[bilal-aamer]

@VictorOdede Could you provide some resources and be a bit more descriptive?

[SrijanSahaySrivastava]

Heya, May i work on this. Could you give some more context related to it

[VictorOdede]

@SrijanSahaySrivastava This is about extending the context window of LLMs. One way to do it is through RoPE scaling:https://arxiv.org/abs/2306.15595

[AfamO]

Could this mentioned here Prompt Tuning be of any assistance?

[lucylililiwang]

Hi, can I please work on this issue? Thank you!

[lucylililiwang]

https://github.com/lucylililiwang/LLM-VM/tree/lucylililiwang-memory_state_mamage

[greydoubt]

Amazing! Are you using a wrapper to enforce history structure? Does it need error handling for bad responses?

[lucylililiwang]

Thank you! @greydoubt! I think I might add the error handling for the bad response. Thank you for suggestion!