AWS Lambda Cold Start times

[domvo]

We're experiencing cold start times of roughly 6-8s with langchain, pinecone and openai imported for a very small retrieval function.

Experiments have shown that importing langchain is responsible for ~90% of this cold start times. Function memory makes no difference.

Any ideas on how to reduce these cold starts? Running on python3.9, native python runtime (no docker) and with 2,048mb RAM.

[PradHolla]

Tried using Provisioned Concurrency for Lambda?