đī¸ Encountering a FailedScheduling Event When Starting an Instance
During the startup process of an instance, if a warning event occurs (such as the FailedScheduling event shown below), it will not affect the instance startup. Simply wait for the instance to start.
đī¸ How to Install Software Offline on H100 Instances
To address the issue of H100 GPU instances being unable to connect to the internet, we provide specialized official images for these instances (with the -h100 suffix). When creating an instance, simply select one of these images, and both apt and pip will be configured to use local mirrors for software and dependency installation, enabling software deployment in an offline environment. The local mirrors currently contain the same package list as the Tsinghua repository.
đī¸ SSH Connection Interruption Issues
In Neolink.AI, the server's default timeout for SSH connections is set to 30 minutes. If you experience frequent disconnections during this period, it may be due to an unstable network connection. To reduce interruptions, it is recommended to switch to a more stable network environment. Additionally, you can use JupyterLab for connections, which provides a more persistent working environment. Even if network issues occur, it helps avoid disruptions to your operations.