Could not find conda environment: train about pipeline HOT 3 CLOSED

It seems that the conda environment wan't configured properly. Did you get any warning when build the train image about conda's environment settings? You may check the conda environment in your train pod by

kubectl exec -it <your-train-pod> -- /bin/bash
$you-pod:~# conda env list

from pipeline.

from https://joinslack.pipeline.ai:

I find out the solutions. In fact, you need to add some environment variables in the training Dockerfile templates. More specifically, the official Dockerfile templates lose the PIPELINE_MODEL_PATH and PIPELINE_MODEL_TRAIN_CONDA_ENV_NAME, so you need to config those two variables according to your environment.

from pipeline.

@un-knight

ok, this is now fixed. make sure you do the following:

pipeline env_registry_sync --tag=1.5.0

pipeline train-server-stop --model-name=mnist --model-tag=v1

pipeline train-server-build --model-name=mnist --model-tag=v1 --model-type=pytorch --model-path=./pytorch/mnist-v1/model/

pipeline train-server-start --model-name=mnist --model-tag=v1 --input-host-path=./pytorch/mnist-v1/model/ --output-host-path=./pytorch/mnist-v1/model/ --training-runs-host-path=./pytorch/mnist-v1/model/ --train-args=""

basically, clean your env, then re-run the training steps here:

https://github.com/PipelineAI/pipeline/blob/master/docs/quickstart/docker/README-training.md

thanks @ericwangqing for your input.

we renamed those variables to PIPELINE_RESOURCE_* from PIPELINE_MODEL_*, but the Docker image wasn't yet pushed.

thanks for helping us track this down. please re-open if there is still an issue. we've updated the docs, as well, to fix some minor typos.

from pipeline.