PromptRunner Interface

PromptRunner: Simplifying Prompt Execution

pg.PromptRunner is a core component of PromptoGen, serving the purpose of efficiently executing prompts created by users. Its prime feature is the ability to run prompts without having to be concerned about the specifics of the LLM (Large Language Model) implementation. This significantly simplifies the creation and management of prompts.

Usage Example

import promptogen as pg

# formatter is an object that formats prompts
formatter = pg.KeyValuePromptFormatter()

# For the creation of a TextLLM instance, refer to the `TextLLM` page
runner = pg.TextLLMPromptRunner(llm=text_llm, formatter=formatter)

# Creating an actual prompt
summarizer = pg.Prompt(
    name="Text Summarizer and Keyword Extractor",
    # ...
)

# Running the prompt
input_value = {
    "text": "In the realm of software engineering, ...",
}
output_value = runner.run_prompt(summarizer, input_value)
print(output_value)

Why is pg.PromptRunner useful?

1. Abstraction: Users can execute prompts without being aware of the specifics of the LLM implementation.
2. Consistency: Makes changes minimal when running the same prompt using different LLMs.
3. Extensibility: Easy to add new prompts or modify existing ones.

Thus, pg.PromptRunner is a potent tool in PromptoGen for executing prompts efficiently and conveniently.

Overview of PromptRunner

Users select the desired TextLLM and PromptFormatter and create a PromptRunner.

By providing the Prompt and InputValue to the PromptRunner, you can obtain the OutputValue.

• Input to PromptRunner: Prompt and InputValue
• Output from PromptRunner: OutputValue

PromptRunner operates in the following sequence:

1. Receives Prompt and InputValue
2. Generates a prompt string
3. Passes the generated prompt string to TextLLM
4. Receives output from TextLLM
5. Converts the output to OutputValue using PromptFormatter
6. Returns OutputValue

pg.PromptRunner Interface

PromptRunner has a run_prompt method. This method receives a prompt and input and returns the output parsed to Value (i.e., dict).

Reference: Abstract class for PromptRunner

class PromptRunner(ABC):
    @abstractmethod
    def run_prompt(self, prompt: Prompt, input_value: Value) -> Value:
        pass

pg.TextLLMPromptRunner Implementation

The execution of prompts using pg.TextLLMPromptRunner is performed as follows:

1. Use the PromptFormatter to format the prompt and input values into text.
2. Generate text using TextLLM.
3. Parse the generated text using PromptFormatter to return a Value.

Reference: Implementation of TextLLMPromptRunner

class TextLLMPromptRunner(PromptRunner):
    # ...

    def __init__(self, llm: TextLLM, formatter: PromptFormatter):
        self.text_llm = llm
        self.formatter = formatter

    def run_prompt(self, prompt: Prompt, input_value: Value) -> Value:
        raw_req = self.formatter.format_prompt(prompt, input_value)
        raw_resp = self.text_llm.generate(raw_req)
        resp = self.formatter.parse(prompt, raw_resp)
        return resp

Usage

Here, we use the OpenAITextLLM class introduced in the TextLLM example for TextLLM.

Moreover, for the prompt to be executed, we use the summarizer prompt created in Prompt.

import promptogen as pg

class OpenAITextLLM(pg.TextLLM):
    # ...(omitted)...

text_llm = OpenAITextLLM("gpt-3.5-turbo")
formatter = pg.KeyValuePromptFormatter()

runner = pg.TextLLMPromptRunner(text_llm, formatter)

input_value = {
    'text': "In the realm of software engineering, ...",
}
resp = runner.run_prompt(summarizer, input_value)
print(resp)
# {'summary': 'Software developers collaborate ...', 'keywords': ['software engineering', 'developers', ...]}