Skip to content

Main Concepts

pipeline

Understanding Pipelines and Workflows

Pipelines and Workflows are both Griptape Structures that execute a series of tasks.

Pipelines

Pipelines are always a sequential series of steps - one task after another until it is finished.

In this course, we're going to be taking a topic to draw, and then execute a consistent series of tasks until we have the image. Every time the Pipeline is run, it will always:

  1. Use an LLM to generate an Image Generation prompt.
    • If the input given is "a butterfly", the result of this task might be a text string "Create a watercolor painting of a butterfly.".
    • This will allow us to ensure the output from this task will always be in the correct format for an Image Generation Driver.
  2. Generate an image
    • Using the output from the previous step, we'll give that to an Image Generation Model
  3. View the image
    • Once the image is generated, we'll want to do something with it. In this case, we can open the image to view it.

The flow of tasks will look like:

graph LR
    A((Get Topic)) --> B(Generate Prompt) --> C(Generate Image) --> D(Show Image)

Workflows

Workflows allow for complex interactions, resembling tree branches.

Workflows are non-sequential, and individual tasks can depend on multiple input tasks. This allows you to create a single task that waits for all other tasks to complete before it can begin.

If we wanted to generate multiple images, perhaps in different styles we'd be able to set up a workflow like the following:

Note

The graph is drawn top to bottom for this example because it's easier to understand the flow of data, but it can be drawn in either direction.

graph TB
    A((Get Topic)) --> B(Generate Prompt 1) --> D(Generate Image 1) --> I(Show Images)
    A --> E(Generate Prompt 2) --> F(Generate Image 2) --> I
    A --> G("Generate Prompt <i>n</i>" ):::dash
    G --> H("Generate Image <i>n</i>"):::dash 
    H --> I
    classDef dash stroke-dasharray: 5 5

Notice how the images can be generated in parallel, but the Show Images task will wait until all its parent tasks are completed. Workflows always require a start task and an end task.

Tasks

Before we dive in and start setting up our own Pipeline, it's important to review the concepts of Tasks. With Griptape, there are many types of tasks you'll be working with, including:

Task Type Description Example
PromptTask General purpose prompting to the LLM. PromptTask("Tell me a story about skateboards")
ToolkitTask Uses Griptape Tools to complete a task with Chain of Thought (CoT) reasoning. ToolkitTask("Summarize griptape.ai", tools=[WebScraperTool()])
ToolTask Similar to ToolkitTask, but only uses a single tool and no CoT. ToolTask("Give me the answer for 52-10", tool=Calculator())
Extraction Tasks Various tasks associated with extracting information from text. See examples in the documentation.
TextSummaryTask Summarizes text very efficiently TextSummaryTask("Imagine this is a massive amount of text.")
TextQueryTask Can be used to query large bodies of text, for example a vector database. See examples in the documentation
PromptImageGenerationTask Can be used to generate images based off a text prompt. PromptImageGenerationTask("watercolor butterfly"), image_generation_driver=image_driver
CodeExecutionTask Can be used to execute code. CodeExecutionTask(on_run=reverse_string)

In this course, we will be focusing mostly on Prompt Tasks, Image Generation Tasks, and Code Execution Tasks. 

# Example PromptTask to create a Image Generation Prompt
#
image_prompt_task = PromptTask(
    "Create a prompt for an image generation engine that will make a watercolor painting of: {{ topic }}",
    context = {                        
        "topic": "butterfly" 
    },
    id="image_prompt_task"   # task id can be referenced by other tasks
)

# This Image Generation task works with the output of the parent task.
#
image_generation_task = PromptImageGenerationTask(
    "{{ parent_output }}", # The output of the parent task
    image_generation_driver=OpenAiImageGenerationDriver(
        model="dall-e-3", api_type="open_ai", image_size="1024x1024"
    ),
    output_dir="./images",
    id="image_generation_task"
)

Next Steps

Let's move on to the next section where we'll create our First Pipeline, and get an understanding of the basics of how parent/child relationships can work.