Skip to content

Latest commit

 

History

History
62 lines (60 loc) · 2.19 KB

README.md

File metadata and controls

62 lines (60 loc) · 2.19 KB
Raw

Natural language processing services

This is a list of several ready to use NLP services, wich i used in practice and built as distinct docker containers, preferrly required Nvidia GPU adapter.

Installation

git clone https://github.com/format37/nlp.git
cd nlp
mv docker-compose.yml.default docker-compose.yml

Open and edit docker-compose.yml
To enable the selected service, set according service replicas parameter:

replicas: 1

Define, wich nvidia gpu adapter used for service, starting from 0:

device_ids: ['0']

Define the model, corresponding to language you need, if available in docker-compose.yml config.
To build server, run:

sudo docker-compose up --build -d

And test the service, with one of examples below

Deeppavlov paraphrase

Docker service, receives two lists of phrases.
Returns a list[n] of int values: 0 or 1, wich defines, is list_a[n] and list_b[n] are paraphrase or not.
Example: paraphrase.ipynb

Deeppavlov sentiment

Docker service, receives one list of phrases.
Returns a list[n] of sentiment categories ['positive', 'neutral', 'negative', 'speech'], for each list[n] phrase.
Example: sentiment.ipynb

Deeppavlov textqa

Docker service, receives two lists:

  • texts
  • questions

Returns two lists:

  • questions
  • answers

Example: textqa.ipynb

Summarus

Docker service, receives one string of Russian text.
Returns one summarized string of text.
Example: summarus.ipynb

Parse unstructured data

Direct openai api request, using text-davinci-002 model, to parse unstructured data.
For example, having a prompt:

Alex: How much you paid for pizza?
Jane: I paid $7 for my piece of pizza
Max: I am not eating that pizza
Alex: Ok, then my piece was $12

Unparsed answer looks like:

|------|-------|
| Alex | $12   |
| Jane | $7    |
| Max  | $0    |

Example: parser.ipynb