How to set the Cache?

[vlavorini]

In Langchain, we can set the LLM Cache:

from langchain.globals import set_llm_cache
set_llm_cache(...)

But I see that just this setting is not sufficient for LangGraph to use the same cache.

Am I doing something wrong, or maybe this is not implemented for langgraph?

[hwchase17]

are you trying to cache LLM calls or ALL of langgraph? If LLM calls, it should work the same (assuming you are using LangChain LLMs). If all of LangGraph - not currently supported, but its on our roadmap. OOC - what is the motivation there?

[vlavorini]

At least the LLM calls, but that does not happen. No records arrive to the cache DB.

I just replicated the first part of the tutorial using different tools (RAG and Pandas), and added the cache.

We want this to speed up the calls, given that the Pandas tool is really slow.

[NajiAboo]

Any update on the Langgraph cache..I am looking for all calls.

[KaiKin-C]

Same question. If I invoke the LLM directly, it uses the cache. However, If I compile a graph with the same LLM inside (I even pass the cache argument in compile() ), the LLM does not use the cache across runs even the input is the same. I wonder why it is the case.

[hinthornw]

https://langchain-ai.github.io/langgraph/concepts/low_level/?h=cache#node-caching
?

[kamaldeepj]

Yes I want to use Redis as replacement for InMemoryCache for Node Caching.