Hi, I was able to make and run my inpainting code in MLCube. It is passing all test cases up until the evaluation. This is the error I get: > Model execution complete 2024-08-06 19:08:09 Kurtlab-Agam mlcube.__main__[274867] INFO Running task = evaluate 2024-08-06 19:08:09 Kurtlab-Agam mlcube.__main__[274867] ERROR DockerRun runner failed to run MLCube. Error occurred while generating mount points for docker run command (task=evaluate): Invalid task: task=evaluate, param=predictions, type=unknown. Type is unknown and unable to identify it (/home/agam/.medperf/predictions/api_medperf_org/e3a36e7c0e5700502d6c2f088b9fc9ca86d062e431c9ad3543a407427a479f23/cdb02f6d6d3b1657dfb744b49a7ef95044bd7188cd10ab444f4e5ba99b6b8cf7). ? Metrics MLCube failed. I understand that the test case does not pass a directory to save the results using --output_path, so my inference code in this case simply returns the predictions in nifti format(nii.gz). Is there something I'm misunderstanding or do we need to explicitly write the outputs to a directory, if so how and where should we do so? Thanks in advance!

Created by Agamdeep Chopra achopra
As a followup, I was able to fix this error. The issue was that I used tab for indenting the outputs line in the yaml file instead of spacebar. Switching to space fixed the error!
Thank you so much!
Hi @achopra , I forwarded your question to one of the MedPerf developers and they have responded with: > The model MLCube should mount an output folder named ?predictions?, and store any assets related to predictions there. From my understanding they?re instead writing directly to a file, which will not live in the expected folder, causing the evaluation MLCube to break.

MLCube test case failing during evaluation page is loading…