External QA (and possibly prod) is using machine.py:latest for word alignment

[pmachapman]

The following ClearML job:

from machine.jobs.build_word_alignment_model import run
args = {
    'model_type': 'thot',
    'engine_id': '693f832a5638247db937eca2',
    'build_id': '693f83305638247db937ecb8',
    'shared_file_uri': 's3://silnlp',
    'shared_file_folder': 'ext-qa/',
    'build_options': '''{}''',
    'clearml': True
}
run(args)

Displayed this in the console:

Running Task 6b0651a2909b4a3697055ef04606bb25 inside docker: ghcr.io/sillsdev/machine.py:latest arguments: ['--env', 'ENV_FOR_DYNACONF=production']

And failed (due to an incompatibility with the latest machine.py image I pushed).

Should the word alignment engine be using the same version of machine.py as SMT and NMT instead of latest?

[Enkidu93]

Yes, it should use the same version SMT uses (which is <NMT version>.cpu_only). Looks like we need to add an additional section to the secrets.yml for the statistical model type. BuildJob__ClearML__2 corresponds to the statistical word alignment engine.

            - name: BuildJob__ClearML__0__Queue
              value: {{ .Values.ClearMLQueue}}
            - name: BuildJob__ClearML__0__DockerImage
              value: {{ .Values.ClearMLDockerImage}}
            - name: BuildJob__ClearML__1__Queue
              value: {{ .Values.ClearMLQueue}}.cpu_only
            - name: BuildJob__ClearML__1__DockerImage
              value: {{ .Values.ClearMLDockerImage}}.cpu_only

If we end up moving to appsettings (#27), we can change it there instead.