[go: up one dir, main page]

Skip to content

This repo contains the code written primarily in Golang for a self-healing large language model (LLM) pipeline that iteratively corrects errors in its own generated code.

License

Notifications You must be signed in to change notification settings

ammarlodhi255/Self-healing-LLM-Pipeline

Repository files navigation

Self-Healing LLM Pipeline

Self-Healing LLM Pipeline uses open-source Large Language Models (LLMs), like llama3.1, to iteratively generate, test, and refine code. It automates the process of identifying and fixing issues in generated code through a feedback loop, minimizing the need for human intervention. This application supports integration with Ollama for model management and API interactions.

  • The LLM validates its generated main code by writing three test cases.
  • Separates the main code and test code, saving them in respective files.
  • The main code is compiled first; if it fails, the errors are fed back to the LLM.
  • If the main code compiles, the test cases are compiled and executed, and if any error within the test file or any failed tests exists, they are returned as feedback.
  • This loop continues until the code:
    • Compiles successfully
    • Passes all three test cases.

Features

  • Iterative Self-Healing Loop: Automatically generates, tests, and refines code until it compiles successfully.
  • Flexible Model Selection: Choose between models like llama3.1, codellama, and codellama:13b.
  • Dual Modes of Operation:
    • Single Prompt: Submit a single prompt and observe the LLM's iterative process.
    • Prompt List: Upload a list of prompts to process sequentially.
  • Detailed Metrics: Tracks elapsed time, number of iterations, and results for each prompt.
  • Persistent Model Management: Integrates with Ollama for managing models and API requests.

Prerequisites

Before setting up the project, ensure the following are installed on your machine:

  • Go 1.19+
  • Docker
  • Ollama (running on the host machine)
  • Access to the required LLM models, such as llama3.1.

Installation

Clone the Repository

git clone https://github.com/your-username/llm-self-healing-pipeline.git
cd llm-self-healing-pipeline

Run the application locally:

To run the application locally without Docker:

  1. Install Go dependencies:
go mod tidy
  1. Run the application:
go run main.go
  1. Access the application at http://localhost:8080.

Build and Run with Docker

docker build -t llm-pipeline .

Run the Docker Container

Mount your host’s Ollama directory to the container for model access:

docker run -d \
    -p 8080:8080 \
    -p 11434:11434 \
    -v /Users/username/.ollama:/root/.ollama \
    --name llm-pipeline \
    llm-pipeline

• p 8080:8080: Exposes the web interface for interacting with the application. • -p 11434:11434: Exposes Ollama’s API for model interactions. • -v /Users/ammarahmed/.ollama:/root/.ollama: Maps the Ollama model directory for persistent access.

Verify the Setup:

  • Access the web interface at http://localhost:8080.
  • Use the Single Prompt or Prompt List mode to interact with the application.

Usage

Single Prompt Mode

  1. Navigate to the Single Prompt section in the UI.
  2. Enter a prompt and select a model from the dropdown.
  3. Click Submit to start the iterative process.
  4. Observe:
    • The generated code.
    • Compiler output.
    • Progress metrics such as iterations and elapsed time.

Prompt List Mode

  1. Navigate to the Prompt List section in the UI.
  2. Upload a .txt file containing prompts, one per line.
  3. Select a model from the dropdown and start processing.
  4. The application will process each prompt sequentially and display results.

Monitoring Metrics

  • Elapsed Time: Total time spent on processing.
  • Iterations: Number of iterations performed for the current prompt.

Technical Details

Project Structure

project-root/
├── Dockerfile                     # Containerization setup
├── main.go                        # Backend Go application
├── static/                        # Frontend assets (CSS, JS)
├── imgs/
├── templates/
│   └── index.html                 # Frontend HTML template
├── go.mod                         # Go module dependencies
├── go.sum                         # Go module checksums
└── modules/                       # Core modules for LLM processing and utilities
    ├── compiler/                  
    ├── compiler_v2/
    │   ├── consts/                # Constants for compiler configurations
    │   ├── go_compiler_v2/        # Go-specific compiler logic
    │   ├── platform/              # Cross-platform handling
    │   ├── rust_compiler_v2/      # Rust-specific compiler logic
    │   └── utils/                 # Utility functions for compiler operations
    ├── database/                  
    │   ├── db.go                  # Database logic and configuration
    │   └── test_db.go             # Unit tests for database interactions
    ├── display-indicator/         
    │   └── indicator.go           # Display indicators and loading animations
    ├── extraction/                
    │   ├── extract.go             # Extraction logic for code from LLM response
    │   └── extract_test.go        # Unit tests for extraction functions
    └── ollama-implementation/     
        ├── ollama.go              # Ollama API interaction logic
        └── ollama_test.go         # Unit tests for Ollama API functions

Screenshots

Main Interface with Prompt Input and Model Selection

Main Interface with Prompt Input and Model Selection

Model Selection Dropdown

Model Selection Dropdown

Code Generation Process with Main Code and Test Code Sections

Code Generation Process with Main Code and Test Code Sections

Detailed Output with Compiler Messages

Detailed Output with Compiler Messages

License

License

This project is licensed under the MIT License. See the LICENSE file for details.

About

This repo contains the code written primarily in Golang for a self-healing large language model (LLM) pipeline that iteratively corrects errors in its own generated code.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published