/o1_at_home

open-webui pipe the aims to replicate the o1 experience

Primary LanguagePythonMIT LicenseMIT

Think-Respond Chain Pipe: o1 at home

Think-Respond Chain Pipe

Description

The o1 at home pipe is a Manifold pipe designed to process user queries using a two-step process:

  1. Thinking Step: Generates internal reasoning based on the query.
  2. Responding Step: Produces a final response using the reasoning.

This pipe is lightweight and doesn't require an additional pipeline server to run.

Setup Instructions

1. Add the Pipe Manifold

  • Navigate to the Admin Panel.
  • Go to the Functions tab and click the + to add the pipe.
  • Copy the code and give it a name and description if prompted.

2. Enable the Pipe Manifold

  • After adding the pipe, enable it to activate it.

3. Customize Settings

  • Access the configuration menu via the settings cog:
    • Select Models: Choose your thinking and response models.
    • Show Reasoning: Decide whether to display the reasoning process.
    • Set Thinking Time: Specify the maximum allowed thinking time (in seconds).
    • Toggle OpenAI API or Ollama: Specify which API to use.

4. Save and Apply

  • Once customized, save your settings. The pipe should now appear in your dropdown list.

Features

  • Error Checking: Notifies you if a model ID is invalid.
  • Reasoning Visibility: Toggle to show or hide reasoning steps.
  • Time Limits: Set maximum reasoning time to suit your needs.

TODO:

Completed

  • Support for OpenAI API: Fully integrated for reasoning and response.
  • Support for Multiple Models: Select different models for thinking and responding.
  • Streaming Responses: Added streaming response support for real-time interaction.

In Progress

  • Support Multiple Models from Multiple APIs: Allow simultaneous use of OpenAI, Ollama, and other APIs.
  • Customizable APIs per Task: Enable model/API selection based on specific tasks.

Future Enhancements

  • Claude Integration: Add support for Anthropic's Claude models.
  • Additional APIs: Expand support for APIs like Cohere, Hugging Face, and others.
  • Enhanced Error Logging: Provide more detailed insights for API issues.
  • Load Balancing: Add logic to distribute requests across APIs/models for better performance.
  • Advanced Reasoning: Incorporate features for context retention across sessions.
  • Interactive Debugging Tools: Allow real-time debugging for reasoning steps.

Notes

  • This is a Manifold pipe, not a pipeline—no extra pipeline server is required.
  • For detailed usage, check the comments in the code.

Author