Platform For AI:Install a PAI-Megatron-Patch image

Deep Learning Containers (DLC) is a cloud-native training platform that supports custom images, distributed training, and multiple frameworks.

DLC supports custom images for PAI-Megatron-Patch. After installation, run large-scale distributed training on multi-GPU servers.

1. Log on to the PAI console.

2. In the left-side navigation pane, click Workspace List. On the Workspace List page, click a workspace.

3. In the left-side navigation pane, choose Model Development and Training > Deep Learning Containers (DLC). Click Create Job.

4. Configure these parameters. For other parameters, see Create a training job.

   • Environment Information: Set Node Image to Image Address. Enter this address:

     pai-image-manage-registry.cn-wulanchabu.cr.aliyuncs.com/pai/pytorch-training:2.0-ubuntu20.04-py3.10-cuda11.8-megatron-patch-llm

   • Resource Information:

     • Set Framework to PyTorch.

     • Job Resource: In Resource Specification column, click , then select a GPU-accelerated node type and specifications.

   

   

5. Click OK.

Install in DSW

Data Science Workshop (DSW) is a cloud-based development environment that integrates JupyterLab and supports custom plug-ins.

DSW supports custom images. After installation, debug PAI-Megatron-Patch training acceleration programs.

1. Log on to the PAI console.

2. In the left-side navigation pane, click Workspace List. On the Workspace List page, click a workspace.

3. In the left-side navigation pane, choose Model Development and Training > Data Science Workshop (DSW). Click Create Instance.

4. Configure these parameters. For other parameters, see Create a DSW instance.

   • Resource Quota: Select Public Resources (Pay-as-you-go).

   • Resource Specification: Click , then select a GPU-accelerated instance specification.

   • Image: Enter this address: pai-image-manage-registry.cn-wulanchabu.cr.aliyuncs.com/pai/pytorch-training:2.0-ubuntu20.04-py3.10-cuda11.8-megatron-patch-llm

   

5. Click OK.