Can I load my Rasa NLU model stored in AWS S3 in Rasa server using a python script?

Question

I have trained a Rasa NLU model for intent classification and entity extraction and uploaded it to a S3 bucket.

Now how do I load it to Rasa Server using a python script?

For example, using a CLI cmd: rasa run --model spacynlu.tar.gz --remote-storage aws I can start the rasa server and load the model.

How do I achieve this using a python script?

Answer 1

A bit late to the party, but here is a solution for rasa 3.2.6.

I'm not sure about your exact use case - whether you want to launch the Rasa server from a Python script, or to load a model and use it for predictions. The solution below is for the later case. However, it should also help for the former use case - in general, you will need to set the environment variables as described below, and then use os.system(server_cmd) for launching the server, where server_cmd is the command you usually use for running the server, for example

server_cmd = "rasa run --model spacynlu.tar.gz --remote-storage aws"

just make sure that the value of --model is the full path in S3 to your model.

Using rasa for parsing messages in a Python script, using a model stored on AWS S3

Rasa is using boto3 for accessing S3, so you need to have it installed. Then you need to set the environment variables used by boto3 for accessing and authenticating. For example:

import os
os.environ["AWS_ENDPOINT_URL"] = "https://s3.eu-west-1.amazonaws.com"
os.environ["AWS_DEFAULT_REGION"] = "eu-west-1"
os.environ["BUCKET_NAME"] = "my-bucket"

You can, of course, set these variables in the shell, just make sure you export them.

For knowing the correct value for AWS_ENDPOINT_URL you can consult AWS documentation at https://docs.aws.amazon.com/general/latest/gr/s3.html.

Now you can initialize an agent

import asyncio
from rasa.core.agent import Agent, load_agent

agent = asyncio.run(load_agent(
    model_path="my/path/to/models/mymodel.tar.gz",
    remote_storage="aws"
))

load_agent is a global function in rasa.core.agent. Note the model_path is the full path to the model, starting from the S3 bucket. Also note the use of asyncio.run - this could probably be done in other ways, for example, for loading multiple agents/models simultaneously.

Once you obtain an agent instance, you can call, for example, the parse_message method on it

result = asyncio.run(agent.parse_message("Hello world!"))

(again, you can use asyncio in other ways.)

You can read the Rasa official documentation at https://rasa.com/docs/rasa/model-storage/#load-model-from-cloud, but it lacks some details.