search

maana-io / q-cli

Custom GraphQL CLI plug-in for the Maana Q Knowledge Platform, including project scaffolding, data uploading, service deployments, and more.
2 stars 4 forks source link

Python version conflicts #74

Open fastflair opened 4 years ago

fastflair commented 4 years ago

Is your feature request related to a problem? Please describe. I'm not able to deploy a model that requires python version 3.7 when my environment has multiple version of python. The local environment (both anaconda and the operating system, Ubuntu 18.04 in WSL2) uses python 3.7 Ubuntu WSL2 was installed with python version 3.8 initially. When I'm in the command line, python --version correctly reports python 3.7.9 and all things perform as expected. When the command graphql mdeploy is run, somehow python 3.8 is invoked: Sending build context to Docker daemon 6.247GB Step 1/11 : FROM tiangolo/uvicorn-gunicorn:python3.8

As a result, the required library tensforflow==1.15 is not installed as it's not supported with python version 3.8.

Describe the solution you'd like It seems like the environment should be built/deployed using the python version that is active?

Describe alternatives you've considered It's not feasible to upgrade to tensorflow2.0, we did, but the revised structures result in exceeding memory limits without downsizing the models.

Additional context Wanting to deploy an extra large gpt2 model (tensorflow 1.15 framework - works with CPUs efficiently on standard hardware with extra large models) trained to generate abstracts from titles and summarize non domain specific scientific articles.

fastflair commented 4 years ago

Resolved it, the setting was in the first line of the dockerfile: FROM tiangolo/uvicorn-gunicorn:python3.8 changed to FROM tiangolo/uvicorn-gunicorn:python3.7