Skip to content

Latest commit

 

History

History
156 lines (110 loc) · 4.5 KB

README.md

File metadata and controls

156 lines (110 loc) · 4.5 KB
Raw

Documentation

Rasa model server with upload endpoint
Version
Simple webserver for externalizing RASA models.

About

You can configure RASA to fetch models from this server in two ways.

Get a specific model

Pointing to a specific model (.tar.gz) and overriding said file when you want the model to change. http://localhost:8080/models/model.tar.gz

Get the latest model

Pointing to a folder (suffixing the url with @latest) containing multiple models (.tar.gz) and getting the latest model sorted by modified date. http://localhost:8080/models/@latest

Upload a model

You can upload models via POST requests that you send during a CI/CD RASA model training job. If you choose a model name that exists on the server, the new one overwrites the old one.

Quick start

Build a Docker image from this repository.

docker build -t rasa-model-server .

Specify your desired configuration and run the container:

docker run -<d|i> --rm \
    -v /host/path/to/models:/app/models \
    -p <host_port>:8080 \
    rasa-model-server

You can stop the container using:

docker stop rasa-model-server

Configuration

You can configure the service via the following environment variables.

Environment Variable Default Value Description
PORT 8080 Port on which to run the webserver.
MODELS_DIR models The absolute or relative location of the models folder.

Examples

Get model

Fetch a model without specifying a If-None-Match header.

curl -s -I 'http://localhost:8080/models/model.tar.gz'

HTTP/1.0 200 OK
Content-Disposition: attachment; filename=model.tar.gz
Content-Length: 6478848
Content-Type: application/x-tar
Last-Modified: Tue, 23 Apr 2019 12:28:43 GMT
Cache-Control: public, max-age=43200
Expires: Fri, 26 Jul 2019 23:42:05 GMT
ETag: "1556022523.364716-6478848-1948524791"
Date: Fri, 26 Jul 2019 11:42:05 GMT
Accept-Ranges: bytes
Server: Werkzeug/0.14.1 Python/3.6.3

Once the model is loaded by RASA, subsequent requests will use the received ETAG to check if the model has been updated.

curl -s -I 'http://localhost:8080/models/model.tar.gz' -H 'If-None-Match: 1556022523.364716-6478848-1948524791'

HTTP/1.0 304 NOT MODIFIED
Content-Disposition: attachment; filename=model.tar.gz
Cache-Control: public, max-age=43200
Expires: Fri, 26 Jul 2019 23:42:48 GMT
ETag: "1556022523.364716-6478848-1948524791"
Date: Fri, 26 Jul 2019 11:42:48 GMT
Accept-Ranges: bytes
Server: Werkzeug/0.14.1 Python/3.6.3

Update the model on the server an the next request will pull the new model upon ETag mismatch.

curl -s -I 'http://localhost:8080/models/model.tar.gz' -H 'If-None-Match: 1556022523.364716-6478848-1948524791'

HTTP/1.0 200 OK
Content-Disposition: attachment; filename=model.tar.gz
Content-Length: 900
Content-Type: application/x-tar
Last-Modified: Sat, 29 Dec 2018 23:17:54 GMT
Cache-Control: public, max-age=43200
Expires: Fri, 26 Jul 2019 23:43:32 GMT
ETag: "1546125474.453404-900-1948524791"
Date: Fri, 26 Jul 2019 11:43:32 GMT
Accept-Ranges: bytes
Server: Werkzeug/0.14.1 Python/3.6.3

Upload model

Upload a model to the server using a POST request.

curl -X POST -F "model=@model.tar.gz" -H "Content-Type: multipart/form-data" "http://localhost:8080/models/model.tar.gz"

Roadmap

  • Add security token to access the server's endpoints.
  • Add a front end that supports all features in app.py
  • Add tests for different requests to check if abort clauses trigger correctly.

Change history

1.2.0

  • Added model upload via POST request Stefan Wellsandt, BIBA - Bremer Institut für Produktion und Logistik GmbH
  • Extended config.py file
  • Updated deprecated parts related to Flask 2.1.0
  • Adding some detailed comments
  • Updated the readme file (e.g. added change history)

1.1.0

  • Several updates by Guilherme Guy
  • Added '..' path check to improve security by Daniel Gabardo

1.0.0

Initial version provided by Victor Rachieru with model download and model index (list)

License

MIT