Large Memory & Time Utilization for loading a Model

Getting Started with Rasa
1

Hi Guys, Hope you all are doing well, i have some queries , i am working on building a model which needs the NLU alone, the records which i use in look up table for slot identification has about 25-30 Lakh data’s in it. and i also have a Custom component. My Pipelines are :

language: “en”

pipeline:

  • name: “WhitespaceTokenizer”
  • name: “RegexFeaturizer”
  • name: “deepPavlov.DeepPavlov”
  • name: “CRFEntityExtractor” features: [ [“low”, “title”, “upper”], [“bias”, “low”, “prefix5”, “prefix2”, “suffix5”, “suffix3”,“suffix2”, “upper”, “title”, “digit”, “pattern”], [“low”, “title”, “upper”] ]
  • name: “EntitySynonymMapper”
  • name: “CountVectorsFeaturizer”
  • name: “EmbeddingIntentClassifier”
  • name: “DucklingHTTPExtractor” url: http://rasa-support timezone: UTC dimensions:
    • time
    • number
    • amount-of-money
    • distance
    • ordinal

policies:

  • name: MemoizationPolicy
  • name: KerasPolicy
  • name: MappingPolicy

version:

rasa -> 1.10.3

My Queries are : -> My model occupies about 8 - 11 GB of RAM, Which is so much for my Machine to Handle is there a way to reduce it. -> While Loading a model it takes about 12 - 15 minutes for the model to load and classify the text.(First Response Alone) -> RASA takes about 10 hours to Train a Model. is there a way to reduce it.

2

Hi prasanth,

8 - 11 GB does feel like a large chunk of memory. How do you measure this? Also, how large is your nlu.md file?

I am unfamiliar with the deepPavlov component, could you list a source where you got it from? Could you also run your model without it and report back if you get a speedup?