This repository compiles prescriptive guidance and code samples demonstrating how to operationalize AlphaFold batch inference using Vertex AI Pipelines.
Apache License 2.0
67
stars
30
forks
source link
Container Image with the New Dockerfile of PR28 causing failure in the Relax step with T4 and A100 #30
After using the Dockerfile introduced by PR28 to create the container image, it causes the relax step failure due to CUDA incompatibility issue when using A100 or T4, even though it's working fine with the Predict step.
If you use L4, however, both Predict & Relax steps are fine. Before a formal fix is worked out, we will add the previous version of Dockerfile as an additional Dockerfile for you to create an extra container image which works with A100 and T4 without the Relax step failure.
chmstimoteo
commented
1 year ago
After using the Dockerfile introduced by PR28 to create the container image, it causes the relax step failure due to CUDA incompatibility issue when using A100 or T4, even though it's working fine with the Predict step.
If you use L4, however, both Predict & Relax steps are fine. Before a formal fix is worked out, we will add the previous version of Dockerfile as an additional Dockerfile for you to create an extra container image which works with A100 and T4 without the Relax step failure.