Syncom is a Python library designed to assist in the creation and execution of pipelines for AI models. It provides a set of classes and functions to define steps in a pipeline, create a pipeline, and run it.
- Pipeline Creation: Define a sequence of steps to be executed in a pipeline.
- Step Definition: Define individual steps in a pipeline, such as data input, logging, and AI model execution.
- Pipeline Execution: Run a defined pipeline and get the result.
- Python 3.7 or later
pydantic- litellm
- instructor
To use this library, you need to define the steps of your pipeline, create the pipeline, and then run it. Here's a basic usage example:
import asyncio
from pydantic import Field
from syncom import (
TypedTemplate,
TemplatePromptStep,
LLMStep,
RunnableConfig,
run_pipeline,
make_pipeline
)
class PingTemplate(TypedTemplate):
template_string = """
# Instructions
Respond to the user command listed below using the following rules.
When the user command is "ping", respond with "pong".
When the user command is "pong", respond with "ping".
# User Command
{{ command }}
"""
command: str = Field(enum=["ping", "pong"])
class PongResult(BaseModel):
command: str = Field(
description="Response command from AI"
enum=["ping", "pong"]
)
# Define the steps
steps = (
TemplatePromptStep(PingTemplate),
LogStep("Ping Template"),
LLMStep(model, PongResult)
)
# Create the pipeline
pipeline = make_pipeline(RunnableConfig(), *steps)
# Run the pipeline
async def main(self):
global pipeline
input_data = {"command": "ping"}
output: PongResult = await run_pipeline(pipeline, input_data)
print(f"Output Command: {output.command}")
if __name__ == "__main__":
asyncio.run(main())This will create a pipeline that generates a prompt from a template and then runs the prompt against GPT-4 resulting in a typed PongResult response.
Contributions are welcome. Please open an issue to discuss your ideas or submit a Pull Request with your changes.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent