TaskD Agent APIs

def init_taskd_agent(config : dict = {}, cls = None) -> bool

Function Description

The user-side code calls this function to initialize TaskD Agent.

Input Parameters

Table 1 Parameter description

Parameter	Type	Description
config	dict:{str : str}	Agent configuration information, including Agent configuration and network configuration. The keys include: `Framework`: Agent framework, currently supports PyTorch and MindSpore `UpstreamAddr`: Upstream IP address on the network side `UpstreamPort`: Upstream port on the network side ServerRank: Agent rank number
cls	Specific instance type	This input parameter is used under the PyTorch framework and is a `SimpleElasticAgent` instance. It is not required for other frameworks.

Return Value

Table 2 Return value description

Return Value Type	Description
bool	Indicates whether the initialization is successful. `True`: Initialization succeeded. `False`: Initialization failed.

Function Description

The user-side code calls this function to start TaskD Agent.

Input Parameters

None

Return Value

Table 1 Return value description

Return Value Type	Description
Varies	The return result is determined by the main execution logic of the Agent under the framework. Agents under different frameworks will have different return results after startup. For example, under the PyTorch framework, `SimpleElasticAgent run()` will return the training result.

Function Description

The user-side code calls this function to register a TaskD Agent callback function.

Input Parameters

Table 1 Parameters

Parameter	Type	Description
operator	str	Key for registering the callback function, such as `START_ALL_WORKER`.
func	callable	The corresponding callback function.

Return Value

Table 2 Return value description

Return Value Type	Description
bool	Indicates whether the registration is successful. `True`: Registration succeed. `False`: Registration failed.