Pointers

Instead of referring to pipelines by ID, you can also use pointers. If you're familiar with Docker tags, this works in a similar way. The pointer "points" to the pipeline, so to speak, and you can use this pointer to specify a pipeline, rather than the pipeline ID.

There are many uses for pointers. A common one is if you have several versions of a pipeline but you want to signify which one is the most stable version. This then allows you to update the pointer to point to a more recent version in the future, without having to rely on using pipeline IDs.

Versioned pointers

Pipelines are automatically versioned under the hood. This is achieved by creating an incremental pointer based on the name of the pipeline. For example, the first time uploading a pipeline named, myuser/gptneo, the pointer myuser/gptneo:v1 will be created behind the scene. If we modify my pipeline and then upload it again using the same name, a new version will be created with pointer myuser/gptneo:v2.

Listing pointers

You can view all the pointers you have available using the following CLI command:

pipeline get pointers

Creating a pointer

You can use the CLI to create a pointer:

pipeline create pointers <pointer_name> <source_pipeline>

Here, pointer_name is the pointer you want to create and source_pipeline can either be a pipeline id or another pointer.

Note that the pointer name must start with your username and pipeline name in the format username/pipeline_name:tag

For example, you can create a latest tag that points to the v2 version of the pipeline above, as follows:

pipeline create pointers myuser/gptneo:latest myuser/gptneo:v2

or using the pipeline ID:

pipeline create pointers myuser/gptneo:latest pipeline_abcd1234

Then, you can make an inference request by using this pointer value:

curl -X POST '<https://www.mystic.ai/v4/runs'>  
--header 'Authorization: Bearer YOUR_TOKEN'  
--header 'Content-Type: application/json'  
--data '{  
	"pipeline": "myuser/gptneo:latest",  
	}  
'