LLM Memory with LangChain (6.3)

[@ManiSaintVictor]

Love the memento analogy! Makes it stick.

[@דניאלכהן-ג6ג]

If I have the budget for both 4070 Super and 4060Ti, which is better? When not dealing with state of the art models, what is better, is 50% more performance better than +4GB Vram? When more than 12GB Vram is needed the the training will always be very long (1Hour+)? I mean, when needing 16Vram is the training will always be Hours long? Is there a scenario when you need 16Vram but the training is very fast (Minutes)? If for both GPUs I will need the cloud for large models, should I get the 4070 Super for not so large models? Please advise.

[@ashvinnihalani8821]

How does the Summary Memory determine what to store there. Is that include in the System prompt or is another model used as a Summarizer?

[@HeatonResearch]

Its another LLM, with a couple of prompts built into the memory class.

[@kaikai7702]

how to add memory in langchain serve？

[@PicaPauDiablo1]

Thank you