Loading page…
Pitch
Description
Comments
GenSphere is a declarative framework to build LLM applications and an open platform to push and pull them. Think of Docker for LLM applications.
Check the medium article here. GenSphere is a declarative framework to build LLM applications and an open platform to push and pull them. Think of Docker for LLM applications.
GenSphere is framework to build LLM applications by declaring tasks and how they connect on YAML files. That unlocks a ton of features, including:
We break any LLM application down to graphs where each node is either a function call, an LLM API call or another graph itself. By doing so, you can inspect (and edit) any application down to its core components. We want to break the cumbersome pile of unnecessary abstractions that some modern frameworks bring, and which have turned LLM application building quite opaque.
Projects are defined by a set of YAML files and associated python functions and schemas. This make it easy to share your application with other developers, and makes it easy for them to use what you built in more complex workflows.
Once you've broken your application down to its core components as described above, you can push it to an open platform (no registration required). A public ID is generated, and your project is now publicly accessible to anyone. You can pull any project by referencing its ID.
You can also check popularity of published projects, as measured by number of pulls they get.
Because you can reference graphs as nodes on parents graphs, GenSphere makes it extremely easy to compose complex applications from simpler reusable components. You can pull projects from the platform and build with them instead of having to work from scratch.
1. Define your workflow with yaml files: Your project is defined by 3 files:
a) yaml file which represents a graph, where each node is a task to be executed, and which can receive inputs from other nodes. Every node is either a python function call (see below), an AI API call or another yaml file itself.
b) a functions.py file, which contains python functions to be executed in the yaml nodes, or to be used as tools for LLMs
c) a schemas.py file which contains pydantic models (that you can use for structured output schema from openAI)
2. Compose complex workflows by nesting graphs: A node can also be another graph itself, and the SDK will take care to resolve dependencies, check consistency and compose a final execution graph.
3. Define your functions and schemas: Create a .py file with all functions you are planning to use in your project, be it as individual nodes in the execution graph or as tools for function calling by LLMs. You can also create a separate .py file with pydantic models if you want to use openAI structured output feature in any given node.. If you are new to pydantic and openAI structured outputs, check this short tutorial.
4. Leverage integration with LangChain and Composio: when doing function calling, you can leverage tools available in both Composio and Langchain. Check the implementation notebook for further details.
5. Visualize your project: GenSphere creates an interactive graphical representation of your project. The SDK renders a graph, and you can inspect every node to see which functions or schemas are associated with it, as well as its parameters and outputs.
6. Execute the workflow: the GenSphere SDK resolves dependencies and executes your workflow.
7. Push to the platform: once ready, your project can be pushed to our open platform. You submit you yaml file and any associated functions or schemas .py files. An ID is generated, and your project becomes publicly accessible.
8. Pull from the platform: you can pull yaml files, functions and schemas from the platform by providing its ID (generated by the creator of the project in the step above). Its then easy to use these components to build more complex LLM applications.
9. Watch your projects grow: you can check popularity of your projects by measuring the number of pulls it received.
pip install gensphere
Import necessary modules
from gensphere.genflow import GenFlow
from gensphere.hub import Hub
import os
Set your OpenAI API key.
os.environ['OPENAI_API_KEY']="PLACE_YOUR_OPENAI_API_KEY"
Pull a project from the platform
hub=Hub()
hub.pull(push_id='2c03079c-0e33-489e-bbbe-777da744d56f',
yaml_filename='simple_examples.yaml',
functions_filename='simple_examples_functions.py',
save_to_disk=True)
This saves simple_examples.yaml and simple_examples_functions.py to your current working diretory. Now you can execute the flow with.
flow=GenFlow('simple_examples.yaml','simple_examples_functions.py')
flow.parse_yaml()
flow.run()
You can access nodes outputs with
flow.outputs
This notebook contains everything you need to know about GenSphere, including yaml syntax, visualizastion, yaml parsing, nesting workflows, etc.
Check our detailed documentation here
[coming soon]
[coming soon]
In this alpha version, only openAI is supported for AI API calls. We are planning to integrate with other major providers.
Graphs in yaml files need to be DAGs. We are planning to allow for cycles, which will unlock full agentic system building using only GenSphere.
Feedback is welcome! Please reach out by email or join our discord server.
We welcome contributions! Please reach out by email and join our Discord server.
This project is licensed under the MIT License - see the LICENSE file for details.
No comments yet.
Sign in to be the first to comment.