YAML Selection

Various YAML examples are provided by cluster scheduling components. You can select an appropriate YAML example based on the used component, processor type, and job type, and make necessary modifications according to actual requirements before using it.

Resource Information Configuration Using Environment Variables

If Atlas A2 training product are used in the current environment, refer to Table 1 to obtain the corresponding YAML example.
Then, modify and adapt the YAML files of the Atlas 800T A2 training server, Atlas 200T A2 Box16 heterogeneous subrack, and A200T A3 Box8 SuperPoD Server based on the parameter description provided in Table 1.
If Atlas training product are used in the current environment, refer to Table 2 to obtain the corresponding YAML example.
Then, modify the YAML files of the servers (with Atlas 300T training cards) based on the YAML file of the Atlas 800 training server and the parameter description provided in Table 1.
If Atlas A3 training product are used in the current environment, refer to Table 3 to obtain the corresponding YAML example.

**Table 1** YAML files supported by Atlas A2 training product
Job Type	Hardware Model	Training Framework	YAML File Name	Description	How to Obtain
AscendJob	Atlas 900 A2 PoD cluster basic unit	TensorFlow	tensorflow_multinodes_acjob_{xxx}b.yaml	A two-server two-processor job is presented in the example file by default.	Select the corresponding training framework and obtain the YAML file. NOTE: You can run the npu-smi info command to query the number in the processor model, which is indicated by the Name field in the returned message. In the following example, the value of {xxx} is 910.
		PyTorch	pytorch_multinodes_acjob_{xxx}b.yaml	A two-server two-processor job is presented in the example file by default.
		MindSpore	mindspore_multinodes_acjob_{xxx}b.yaml	A two-server 16-processor job is presented in the example file by default.
		TensorFlow	tensorflow_standalone_acjob_{xxx}b.yaml	A single-server single-processor job is presented in the example file by default.
		MindSpore	mindspore_standalone_acjob_{xxx}b.yaml
		PyTorch	pytorch_standalone_acjob_{xxx}b.yaml
		PyTorch	pytorch_multinodes_acjob_{xxx}b_with_ranktable.yaml	A single-server two-processor job is presented in the example file by default. Use Ascend Operator to generate a RankTable file.

**Table 2** YAML files supported by Atlas training product
Job Type	Hardware Model	Training Framework	YAML File Name	Description	How to Obtain
AscendJob	Atlas 800 training server	TensorFlow	tensorflow_multinodes_acjob.yaml	A two-server eight-processor job is presented in the example file by default.	Select the corresponding training framework and obtain the YAML file.
		PyTorch	pytorch_multinodes_acjob.yaml	A two-server 16-processor job is presented in the example file by default.
		MindSpore	mindspore_multinodes_acjob.yaml	A two-server eight-processor job is presented in the example file by default. NOTE: To deliver a single-server eight-processor MindSpore job, change the value of minAvailable in mindspore_multinodes_acjob.yaml to 2 and the value of replicas in Worker to 1.
		TensorFlow	tensorflow_standalone_acjob.yaml	A single-server single-processor job is presented in the example file by default.
		PyTorch	pytorch_standalone_acjob.yaml
		MindSpore	mindspore_standalone_acjob.yaml

**Table 3** YAML files supported by Atlas A3 training product
Job Type	Hardware Model	Training Framework	YAML File Name	Description	How to Obtain
AscendJob	Atlas 900 A3 SuperPoD	TensorFlow	tensorflow_standalone_acjob_super_pod.yaml	A single-server single-processor job is presented in the example file by default.	Select the corresponding training framework and obtain the YAML file.
		PyTorch	pytorch_standalone_acjob_super_pod.yaml	A single-server 16-processor job is presented in the example file by default.
		MindSpore	mindspore_standalone_acjob_super_pod.yaml	A two-server 16-processor job is presented in the example file by default.

Resource Information Configuration Using Configuration Files

If Atlas A2 training product are used in the current environment, refer to Table 4 to obtain the corresponding YAML example.
Then, modify the YAML files of Atlas 800T A2 training server, Atlas 200T A2 Box16 heterogeneous subrack, and A200T A3 Box8 SuperPoD Server based on the parameter description provided in Table 2.
If Atlas training product are used in the current environment, refer to Table 5 to obtain the corresponding YAML example.

**Table 4** YAML files supported by Atlas A2 training product
Job Type	Hardware Model	Training Framework	YAML File Name	Description	How to Obtain
VolcanoJob	Atlas 900 A2 PoD cluster basic unit	TensorFlow	a800_tensorflow_vcjob.yaml	A single-server 16-processor job is presented in the example file by default.	YAML
		PyTorch	a800_pytorch_vcjob.yaml
		MindSpore	a800_mindspore_vcjob.yaml
Deployment	Atlas 900 A2 PoD cluster basic unit	TensorFlow	a800_tensorflow_deployment.yaml	A single-server 16-processor job is presented in the example file by default.	YAML
		PyTorch	a800_pytorch_deployment.yaml
		MindSpore	a800_mindspore_deployment.yaml

**Table 5** YAML files supported by Atlas training product
Job Type	Hardware Model	Training Framework	YAML File Name	Description	How to Obtain
VolcanoJob	Atlas 800 training server	TensorFlow	a800_tensorflow_vcjob.yaml	A single-server eight-processor job is presented in the example file by default.	YAML
		PyTorch	a800_pytorch_vcjob.yaml
		MindSpore	a800_mindspore_vcjob.yaml
	Server (with Atlas 300T training cards)	TensorFlow	a300t_tensorflow_vcjob.yaml	A single-server single-processor job is presented in the example file by default.
		PyTorch	a300t_pytorch_vcjob.yaml
		MindSpore	a300t_mindspore_vcjob.yaml
Deployment	Atlas 800 training server	TensorFlow	a800_tensorflow_deployment.yaml	A single-server eight-processor job is presented in the example file by default.
		PyTorch	a800_pytorch_deployment.yaml
		MindSpore	a800_mindspore_deployment.yaml
	Server (with Atlas 300T training cards)	TensorFlow	a300t_tensorflow_deployment.yaml	A single-server single-processor job is presented in the example file by default.
		PyTorch	a300t_pytorch_deployment.yaml	A single-server eight-processor job is presented in the example file by default.
		MindSpore	a300t_mindspore_deployment.yaml	A single-server single-processor job is presented in the example file by default.

Parent topic: Preparation of Job YAML Files