Dependency Installation
Installing the Rec SDK Torch requires the following environment dependencies and operations. Refer to Table 1 to prepare the installation environment.
Dependency Name/Operation |
Recommended Version |
How to Obtain |
|---|---|---|
Driver and firmware of Ascend hardware products |
Ascend HDK 25.1.RC1 |
Click here. In the Select Resource area on the left, filter the required software packages, confirm the version, and download the software packages. For driver and firmware installation details, see the Driver and Firmware Installation and Upgrade Guide that matches your hardware product. |
Ascend Docker Runtime |
MindCluster 7.1.RC1 |
For details, see "Installation" > "Installation and Deployment" in MindCluster Cluster Scheduling User Guide. |
Device NIC configuration |
- |
Refer to the "Configuring a Training Node" in the Ascend Training Solution Networking Guide. Use HCCN Tool to configure the device IP addresses of NPU network ports. |
CANN software package |
CANN 8.5.0 |
Click here. In the Select Resource area on the left, filter the required software packages, confirm the version, and download the software packages. Obtain the software and perform installation in a container by referring to the "Installing CANN" in CANN Software Installation Guide: Ascend-cann-toolkit_{version}_linux-{arch}.run and Ascend-cann-{chip_type}-ops_{version}_linux-{arch}.run. |
PyTorch Ascend adaptation plugin |
2.6.0/2.7.1 |
Install the plugin of the corresponding version based on the mapping version.
|
For open-source and third-party software integrated by users, track the vulnerabilities and issues in the corresponding community and fix them in a timely manner. You can confirm the known vulnerabilities of the corresponding open-source software version on the CVE official website, and fix the vulnerabilities through version upgrade or patch package update.