Dask on Ray for DaskGraphRunner: Serialization of GraphNode class

[souvikg10 (Souvik Ghosh)]

I think the caching still creates SQL objects. what you can do is make the cache size env var as 0 and try. I think this file is definitely creating the cache object even if you force retraining. so every Node in the GraphPipeline becomes sort of an SQL object. Can you check if you disable cache using the env var as mentioned below, does it still creates a database?

Disable cache folder while Rasa train

Hi @souvikg10, you can set a maximum size for the cache via an environment variable RASA_MAX_CACHE_SIZE (see here). Setting it to 0 will prevent the cache being created (via this function). So RASA_MAX_CACHE_SIZE=0 rasa train should give what you want.

RasaHQ/rasa/blob/main/rasa/engine/caching.py#L37