are somewhere in the pipeline the answers cached?

[touhi99]

Hi, I am currently working on a RAG pipeline with haystack and in the developing stage where I am exploring different options. But i found out the results of the N no. of questions which I am evaluating being cached somewhere probably (its always giving the same exact output, even I am changing parameters of the model temperature, top-k, even prompt text)

My pipeline includes the components:

rag.add_component("text_embedder", text_embedder)
rag.add_component("retriever", retriever)
rag.add_component("prompt_builder", prompt_builder)
rag.add_component("llm", generator)
rag.add_component("answer_builder", answer_builder)
rag.add_component("ranker", ranker)

I am loading generator from vllm models. It is helpful to cache output, but are there some way to control it during the debugging stage? Thanks

[mrm1001]

Hi there @touhi99 , would you mind sharing a bit more about the code you're running? How are you currently changing the parameters of your pipeline? Is it through the run function?

[touhi99]

It was not the pipeline with the issue, but rather from jupyter lab and using seeding. I am closing it.