PromptoGen¶

Bridging LLMs and Python Seamlessly.

Documentation: https://promptogen.zawakin.dev

Source Code: https://github.com/zawakin/promptogen

For the Quick Start Guide, click here.

English | 日本語

PromptoGen: Bridging LLMs and Python Seamlessly.¶

Vision¶

"Seamlessly bridge the gap between LLM and Python, ensuring efficient, future-ready communication."

Key Challenges in LLM Libraries:¶

Lack of an ecosystem for prompt engineering., making prompt creation and sharing difficult.
Strong dependency on specific LLM versions, making them vulnerable to LLM updates.
Complex implementations, hindering customization.

Why Choose PromptoGen?¶

Seamless Conversion between LLM I/O and Python Objects: Streamlining LLM interactions.
Flexible & Unique Interface: Guaranteeing user customizability and extensibility.
Future-Proof Design: Stay ahead with reduced dependency on LLM evolutions.

Compared to Other Libraries: Many are tied to specific LLM versions, lacking the adaptability that PromptoGen offers. With a dependency only on the Pydantic data class library, PromptoGen serves as the ideal bridge between LLM strings and Python objects.

Core Features:¶

Prompt Data Class: Standardizing LLM communication and supporting prompt engineering.
TextLLM Interface: Independence from LLM specifics.
PromptFormatter Interface: High customizability for users.

Benefits:¶

Modular & Extendable: Flexibly mix, match, and add custom components.
Future-Proof: Stand strong against new model updates.
Maintainability: Ensuring easy debugging and minimal adjustments for different LLMs.

Unsupported Features:¶

Direct LLM Communication: We prioritize efficient interfacing over direct LLM conversations.
Prompt Version Management: To keep things streamlined, we avoid adding versioning complexities.
Specific LLM Optimization: Our focus is on adaptability across LLMs rather than optimizing for any single one.

Learn More¶

Dive deep into the About PromptoGen for a comprehensive understanding.

Requirements¶

Python 3.8 or higher

Installation¶

pip install promptogen

Importing¶

import promptogen as pg

How to Use¶

TextLLM: Flexible LLM Integration¶

Through pg.TextLLM, PromptoGen achieves collaboration with a variety of large-scale language models (LLM).

Implementation example of TextLLM interface

import promptogen as pg

class YourTextLLM(pg.TextLLM):
    def __init__(self, model: str):
        self.model = model

    def generate(self, text: str) -> str:
        return generate_by_your_text_llm(text, self.model)

text_llm = YourTextLLM(model="your-model")

By adopting this interface, PromptoGen can seamlessly incorporate different LLMs and their versions. Users can utilize various LLMs in a consistent manner regardless of the specific LLM.

Creating your Prompt¶

import promptogen as pg

summarizer = pg.Prompt(
    name="Text Summarizer and Keyword Extractor",
    description="Summarize text and extract keywords.",
    input_parameters=[
        pg.ParameterInfo(name="text", description="Text to be summarized"),
    ],
    output_parameters=[
        pg.ParameterInfo(name="summary", description="Summarized text"),
        pg.ParameterInfo(name="keywords", description="Extracted keywords from the text"),
    ],
    template=pg.IOExample(
        input={'text': "This is a sample text for summarization."},
        output={
            'summary': "This is a summary of the text.",
            'keywords': ["sample", "text", "summarization"],
        },
    ),
    examples=[
        pg.IOExample(
            input={
                'text': "One sunny afternoon, a group of friends opted to meet at the nearby park to indulge in various sports and activities. They engaged in soccer, badminton, and basketball, reveling in the joy of camaraderie and crafting unforgettable moments together."},
            output={
                'summary': "A bunch of friends relished an afternoon of sports and bonding at a neighborhood park.",
                'keywords': ["friends", "park", "sports", "moments"],
            },
        )
    ],
)

Saving the Prompt¶

summarizer.to_json_file("summarizer.json")

Loading the Prompt¶

import promptogen as pg

summarizer = pg.Prompt.from_json_file("summarizer.json")

Format your Prompt as a String¶

To send an instance of the Prompt class to the LLM (Large Language Model) in actual use, you need to convert it into a string. With PromptoGen, you can use pg.PromptFormatter to convert the prompt into a string in any desired format.

Basic Structure of Prompt Format¶

The basic structure of the prompt format is as follows:

DescriptionInput ParametersOutput ParametersTemplateFew-shot ExamplesInput Value

Prompt Description

<Prompt Description>

Input Parameters:
  - <Description of Input Parameter 1>
  - <Description of Input Parameter 2>
  - ...

Output Parameters:
    - <Description of Output Parameter 1>
    - <Description of Output Parameter 2>
    - ...

Template:
Input:
<Name of Input Parameter 1>: <Example of Input Parameter 1>
<Name of Input Parameter 2>: <Example of Input Parameter 2>
...
Output:
<Name of Output Parameter 1>: <Example of Output Parameter 1>
<Name of Output Parameter 2>: <Example of Output Parameter 2>
...

Example 1:
Input:
<Name of Input Parameter 1>: <Value of Input Parameter 1>
<Name of Input Parameter 2>: <Value of Input Parameter 2>
...
Output:
<Name of Output Parameter 1>: <Value of Output Parameter 1>
<Name of Output Parameter 2>: <Value of Output Parameter 2>
...

...

--------

Input:
<Name of Input Parameter 1>: <Value of Input Parameter 1>
<Name of Input Parameter 2>: <Value of Input Parameter 2>
...
Output:

Input Parameters

<Prompt Description>

Input Parameters:
  - <Description of Input Parameter 1>
  - <Description of Input Parameter 2>
  - ...

Output Parameters:
    - <Description of Output Parameter 1>
    - <Description of Output Parameter 2>
    - ...

Template:
Input:
<Name of Input Parameter 1>: <Example of Input Parameter 1>
<Name of Input Parameter 2>: <Example of Input Parameter 2>
...
Output:
<Name of Output Parameter 1>: <Example of Output Parameter 1>
<Name of Output Parameter 2>: <Example of Output Parameter 2>
...

Example 1:
Input:
<Name of Input Parameter 1>: <Value of Input Parameter 1>
<Name of Input Parameter 2>: <Value of Input Parameter 2>
...
Output:
<Name of Output Parameter 1>: <Value of Output Parameter 1>
<Name of Output Parameter 2>: <Value of Output Parameter 2>
...

...

--------

Input:
<Name of Input Parameter 1>: <Value of Input Parameter 1>
<Name of Input Parameter 2>: <Value of Input Parameter 2>
...
Output:

Output Parameters

<Prompt Description>

Input Parameters:
  - <Description of Input Parameter 1>
  - <Description of Input Parameter 2>
  - ...

Output Parameters:
    - <Description of Output Parameter 1>
    - <Description of Output Parameter 2>
    - ...

Template:
Input:
<Name of Input Parameter 1>: <Example of Input Parameter 1>
<Name of Input Parameter 2>: <Example of Input Parameter 2>
...
Output:
<Name of Output Parameter 1>: <Example of Output Parameter 1>
<Name of Output Parameter 2>: <Example of Output Parameter 2>
...

Example 1:
Input:
<Name of Input Parameter 1>: <Value of Input Parameter 1>
<Name of Input Parameter 2>: <Value of Input Parameter 2>
...
Output:
<Name of Output Parameter 1>: <Value of Output Parameter 1>
<Name of Output Parameter 2>: <Value of Output Parameter 2>
...

...

--------

Input:
<Name of Input Parameter 1>: <Value of Input Parameter 1>
<Name of Input Parameter 2>: <Value of Input Parameter 2>
...
Output:

Template

<Prompt Description>

Input Parameters:
  - <Description of Input Parameter 1>
  - <Description of Input Parameter 2>
  - ...

Output Parameters:
    - <Description of Output Parameter 1>
    - <Description of Output Parameter 2>
    - ...

Template:
Input:
<Name of Input Parameter 1>: <Example of Input Parameter 1>
<Name of Input Parameter 2>: <Example of Input Parameter 2>
...
Output:
<Name of Output Parameter 1>: <Example of Output Parameter 1>
<Name of Output Parameter 2>: <Example of Output Parameter 2>
...

Example 1:
Input:
<Name of Input Parameter 1>: <Value of Input Parameter 1>
<Name of Input Parameter 2>: <Value of Input Parameter 2>
...
Output:
<Name of Output Parameter 1>: <Value of Output Parameter 1>
<Name of Output Parameter 2>: <Value of Output Parameter 2>
...

...

--------

Input:
<Name of Input Parameter 1>: <Value of Input Parameter 1>
<Name of Input Parameter 2>: <Value of Input Parameter 2>
...
Output:

Few-shot Examples

<Prompt Description>

Input Parameters:
  - <Description of Input Parameter 1>
  - <Description of Input Parameter 2>
  - ...

Output Parameters:
    - <Description of Output Parameter 1>
    - <Description of Output Parameter 2>
    - ...

Template:
Input:
<Name of Input Parameter 1>: <Example of Input Parameter 1>
<Name of Input Parameter 2>: <Example of Input Parameter 2>
...
Output:
<Name of Output Parameter 1>: <Example of Output Parameter 1>
<Name of Output Parameter 2>: <Example of Output Parameter 2>
...

Example 1:
Input:
<Name of Input Parameter 1>: <Value of Input Parameter 1>
<Name of Input Parameter 2>: <Value of Input Parameter 2>
...
Output:
<Name of Output Parameter 1>: <Value of Output Parameter 1>
<Name of Output Parameter 2>: <Value of Output Parameter 2>
...

...

--------

Input:
<Name of Input Parameter 1>: <Value of Input Parameter 1>
<Name of Input Parameter 2>: <Value of Input Parameter 2>
...
Output:

Input Value

<Prompt Description>

Input Parameters:
  - <Description of Input Parameter 1>
  - <Description of Input Parameter 2>
  - ...

Output Parameters:
    - <Description of Output Parameter 1>
    - <Description of Output Parameter 2>
    - ...

Template:
Input:
<Name of Input Parameter 1>: <Example of Input Parameter 1>
<Name of Input Parameter 2>: <Example of Input Parameter 2>
...
Output:
<Name of Output Parameter 1>: <Example of Output Parameter 1>
<Name of Output Parameter 2>: <Example of Output Parameter 2>
...

Example 1:
Input:
<Name of Input Parameter 1>: <Value of Input Parameter 1>
<Name of Input Parameter 2>: <Value of Input Parameter 2>
...
Output:
<Name of Output Parameter 1>: <Value of Output Parameter 1>
<Name of Output Parameter 2>: <Value of Output Parameter 2>
...

...

--------

Input:
<Name of Input Parameter 1>: <Value of Input Parameter 1>
<Name of Input Parameter 2>: <Value of Input Parameter 2>
...
Output:

Various Formatting Styles¶

PromptoGen supports various formatting styles:

KeyValue Format: key: value
JSON Format: {"key": "value"}
etc.

To use the key: value format for the prompt, use the pg.KeyValuePromptFormatter. Whether to display parameter information or templates can be set with PromptFormatterConfig.

By using the formatter.format_prompt method, you can convert the prompt and its corresponding input into a string.

KeyValue Format for PromptJSON Format for Prompt

import promptogen as pg

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

formatter = pg.KeyValuePromptFormatter()

input_value = {
    "text": "In the realm of software engineering, ...",
}
print(formatter.format_prompt(summarizer, input_value))

import promptogen as pg

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

formatter = pg.JsonPromptFormatter()

input_value = {
    "text": "In the realm of software engineering, ...",
}
print(formatter.format_prompt(summarizer, input_value))

KeyValue Format for PromptJSON Format for Prompt

Console Output

Summarize text and extract keywords.

Input Parameters:
  - text: Text to summarize

Output Parameters:
  - summary: Summary of text
  - keywords: Keywords extracted from text

Template:
Input:
text: "This is a sample text to summarize."
Output:
summary: "This is a summary of the text."
keywords: ['sample', 'text', 'summarize']

Example 1:
Input:
text: "One sunny afternoon, a group of friends decided to gather at the nearby park to engage in various games and activities. They played soccer, badminton, and basketball, laughing and enjoying each other's company while creating unforgettable memories together."
Output:
summary: "A group of friends enjoyed an afternoon playing sports and making memories at a local park."
keywords: ['friends', 'park', 'sports', 'memories']

--------

Input:
text: "In the realm of software engineering, ..."
Output:

Console Output

Summarize text and extract keywords.

Output a JSON-formatted string without outputting any other strings.
Be careful with the order of brackets in the json.

Input Parameters:
  - text: Text to summarize

Output Parameters:
  - summary: Summary of text
  - keywords: Keywords extracted from text

Template:
Input:
```json
{
 "text": "This is a sample text to summarize."
}```
Output:
```json
{
 "summary": "This is a summary of the text.",
 "keywords": [
  "sample",
  "text",
  "summarize"
 ]
}```

Example 1:
Input:
```json
{
 "text": "One sunny afternoon, a group of friends decided to gather at the nearby park to engage in various games and activities. They played soccer, badminton, and basketball, laughing and enjoying each other's company while creating unforgettable memories together."
}```
Output:
```json
{
 "summary": "A group of friends enjoyed an afternoon playing sports and making memories at a local park.",
 "keywords": [
  "friends",
  "park",
  "sports",
  "memories"
 ]
}```

--------

Input:
```json
{
 "text": "In the realm of software engineering, ..."
}```
Output:

Parsing Outputs from Large Language Models¶

After receiving the prompt string as input, you obtain an output from a large language model (like GPT-3.5, GPT-4).

KeyValue Format for PromptJSON Format for Prompt

LLM Output

summary: "This is a summary of the text."
keywords: ['sample', 'text', 'summarize']

LLM Output

```json
{
    "summary": "This is a summary of the text.",
    "keywords": ["sample", "text", "summarize"]
}
```

You can parse this output as:

KeyValue Format for PromptJSON Format for Prompt

import promptogen as pg

formatter = pg.KeyValuePromptFormatter()

raw_resp = """summary: "This is a summary of the text."
keywords: ['sample', 'text', 'summarize']"""
summarized_resp = formatter.parse(summarizer, raw_resp)
print(summarized_resp)

Console Output

{'summary': 'This is a summary of the text.', 'keywords': ['sample', 'text', 'summarize']}

import promptogen as pg

formatter = pg.JsonPromptFormatter()

raw_resp = """```json
{
    "summary": "This is a summary of the text.",
    "keywords": ["sample", "text", "summarize"]
}
```"""
summarized_resp = formatter.parse(summarizer, raw_resp)
print(summarized_resp)

Console Output

{'summary': 'This is a summary of the text.', 'keywords': ['sample', 'text', 'summarize']}

PromptRunner: Execute Prompts Efficiently¶

pg.PromptRunner supports the execution of prompts simply and efficiently.

Prompt execution using PromptRunner

import promptogen as pg

# Prepare an LLM that implements the `pg.TextLLM` interface
text_llm = YourTextLLM(model="your-model")

formatter = pg.KeyValuePromptFormatter()
runner = pg.TextLLMPromptRunner(llm=text_llm, formatter=formatter)

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

input_value = {
    "text": "In the realm of software engineering, ...",
}

output_value = runner.run_prompt(summarizer, input_value)
print(output_value)

Advantages of this tool:

Abstraction: Users can execute prompts without being aware of the specific LLM implementation.
Consistency: Changes are minimized when executing the same prompt with different LLMs.
Extensibility: Adding new prompts or modifying existing ones is easy.

pg.PromptRunner is a key tool for making prompt execution more intuitive and efficient using PromptoGen.

Quick Start Guide¶

Please refer to the Quick Start Guide.

Application Examples¶

Refer to Application Examples.

Dependent Libraries¶

PromptoGen only depends on Pydantic to define the data class.

Limitations¶

With updates to PromptoGen, compatibility with prompts outputted in JSON may be lost.
The large language models tested for operation are OpenAI's gpt-3.5-turbo, gpt-4, and Meta's Llama 2. Other large language models have not been tested for operation. In particular, there may be cases where the parser does not work correctly, so please be cautious.

Contribution¶

Bug reports, proposals for new features, pull requests, etc., are all welcome! For more details, please see Contribution.

License¶

MIT License