Author: Teng Fu
It is for Bert-QA development purpose, it consists following feature:
-
fastAPI, gunicorn and uvicorn for API development
-
PyTorch 1.6.0
-
Cuda 10.1 and cuDNN 7
-
Huggingface Transformer library
This allows huggingface BERT models able to:
-
execute with GPU-enabled
-
provide REST API service through fastAPI
This image is based on:
pytorch-cuda image is the main base image. In this main base image, the Python version is 3.8.5.
Make sure the nvidia driver is >=R418
sudo docker run -it --runtime=nvidia huggingface-torch-gpu-fastapi:0.1 sh