How can I execute allennlp in codalab?

[minstar]

Hi, I'm stuck in uploading our submission files. The main issue is that codalab is compatible with python version 2.7, however, allennlp is compatible with python version 3.7. So, I really tried to execute serve.py, mrqa_official_eval.py and predict_server.py on codalab in various ways but I failed.. Can anyone help me?

[robinjia]

Try using the --request_docker_image flag in cl run to request a different docker image. For our baseline models, we use the docker image robinjia:mrqa:0.1 (https://hub.docker.com/r/robinjia/mrqa). Be sure to also run your commands with python3.6. You can also use your own custom docker image that you upload to dockerhub.

[minstar]

Thank you for replying, I'll try it!!

[vishalanand]

@minstar Can you share if you got it running? And what was the command you used

[minstar]

I tried command as below,

cl run data_dir:mrqa-dev-data :predict_server.py :mrqa_official_eval.py :bert-cache :baseline "PYTORCH_PRETRAINED_BERT_CACHE=bert-cache python3.6 baseline/serve.py baseline/model.tar.gz 8888 --cude_device 0 & python3.6 predict_server.py <(cat data_dir/*.jsonl) predictions.json 8888; files=$(ls data_dir/*.jsonl); for data_file in ${files}; do base=$(echo ${data_file} | cut -d '/' -f2); python3.6 mrqa_official_eval.py ${data_file} predictions.json > eval_${base::-1}; done" -n run-predictions --request-docker-image robinjia/mrqa:0.1 --request-priority 1 --request_gpus 1

but I don't know why the server doesn't work in gpu version in my case.