Skip to content

Latest commit

 

History

History
40 lines (27 loc) · 1.89 KB

README.md

File metadata and controls

40 lines (27 loc) · 1.89 KB
Raw

NMT API

Implements the NMT API for NeMo AAYN Base models. For more details about building such models, see the official NVIDIA NeMo documentation and NVIDIA NeMo GitHub.

The API provides two endpoints /api/healthcheck, to retrieve the service status, and /api/translate to request a translation. The service accepts either a single string or list of strings. The result will be in the same format as the request, either as a single string or list of strings. The maximal accepted text length is 5000c. Note that transcription of one 5000c text block on cpu will take advantage of all available cores, consume up to 3GB RAM and may take ~200s (on a system with 24 vCPU).

Prerequisites

  • docker >= 20.10.17
  • docker compose >= 2.6.0
  • NeMo model and model.info

Model.info

The expected format for model.info is:

language_pair: # source and target two-letter ISO 639-1 Language Code, lowercase, eg. slen or ensl
domain: # model domain
version: # model version
info:
  build: # build time in YYYYMMDD-HHSS format
  framework: nemo:aayn:base
  ... # aditional info, optional
features: # optional
  ... # information about special features

The NeMo model file is expected in the same folder, named as aayn_base.nemo.

The Neural Machine Translation model for language pair SL-EN developed as part of work package 4 of the Development of Slovene in a Digital Environment, RSDO, project (https://www.slovenscina.eu/en/machine-translation), can be downloaded from http://hdl.handle.net/11356/1736.

Deployment

Run docker compose up -d to deploy on cpu or docker compose -f docker-compose.yml -f docker-compose.gpu.yml up -d to run on gpu.

Approximate memory consumption for cpu deployment

  • 5GB RAM for service and models
  • 3GB RAM per 5000c request