swiss-ai-center/digit-recognition-service

Service that recognizes a digit in an image using mnist trained model.

digit-recognition-service's Issues

Make usage of updated workflow with environment variables

Add export variable for cuda path

In case the model creation fails on on a CUDA enabled machine:

libdevice not found at ./libdevice.10.bc
         [[{{node StatefulPartitionedCall}}]] [Op:__inference_one_step_on_iterator_1668]
ERROR: failed to reproduce 'train': failed to run: python src/create_model.py --dataset data/train.txt, exited with 1

Add the following workaround to documentation:

export XLA_FLAGS=--xla_gpu_cuda_data_dir=/usr/lib/cuda

Ensure cuda is actually installed (add a simlink to /usr/lib/cuda, f.e. from /opt/cuda if needed).

Update service for production

Add CML to digit_recognition service

Add CML to digit_recognition service.

Improve the local development experience

Update version of packages

Related to swiss-ai-center/core-engine#167

Add docs_url field to service description

Integrate MLOps tools for digit_recognition machine learning backend

Problem

At the moment, the digit_recognition machine learning backend uses a model that is stored on DVC with MinIO.

Steps to resolve the problem

Re-set up the tools to allow the model to be built and stored on GitHub Docker registry.