Kiln: Simple LLM model fine-tuning and data synthesis tool, 0 code base to fine-tune your own small models

General Introduction

Kiln is an open source tool focusing on fine-tuning of Large Language Models (LLMs), synthetic data generation and dataset collaboration. It provides an intuitive desktop application with support for Windows, MacOS, and Linux systems that enables users to fine-tune models such as Llama, GPT4o, and Mixtral with zero code and automate serverless deployments.Kiln also supports the generation of training data through an interactive visualization tool that provides Git-based version control for easy team Kiln also supports training data generation through interactive visualization tools and provides Git-based version control for team collaboration on structured data. Its open Python library and OpenAPI REST API make it easy for developers to integrate Kiln datasets into their workflows.

Function List

• Intuitive desktop application: Supports Windows, MacOS and Linux systems with one-click installation and intuitive design.
• Zero-code fine-tuning: Supports fine-tuning of models such as Llama, GPT4o, and Mixtral with automatic serverless deployment.
• Synthetic data generation: Generate training data through interactive visualization tools.
• Teamwork: Git-based version control for team members to collaborate on datasets.
• Tip Generation: Automatically generate prompts from the data, including chainthink, under-sampling, and multi-sample prompts.
• Extensive model and provider support: Support for Ollama, OpenAI, OpenRouter, Fireworks, Groq, AWS and more.
• Open Source Libraries and APIs: Provides the MIT open source Python library and the OpenAPI REST API.
• privacy first: User data is completely private, with support for local operation and self-contained API keys.
• Structured data support: Building JSON-enabled AI tasks.
• free of charge: Desktop applications are free and open source libraries are open.

Using Help

Installation process

1. Download the application: Visit the Kiln GitHub page and select the appropriate installer download for your operating system.
2. Installation of the application::
   • Windows (computer): Run the downloaded .exe file and follow the installation wizard to complete the installation.
   • MacOS: Download the .dmg file, open it and drag Kiln to the Applications folder.
   • Linux: Download the .tar.gz file, unzip it and run the install script.

Guidelines for use

1. Launching the application: After the installation is complete, open the Kiln desktop application.
2. fine-tuned model::
   • Select the "Fine tuning" function module.
   • Select the model to be fine-tuned (e.g. Llama, GPT4o, Mixtral).
   • Upload training data or create a dataset using Kiln's synthetic data generation tool.
   • Configure the fine-tuning parameters and click "Start Fine-Tuning".
   • Once the fine-tuning is complete, the model is automatically deployed with no additional action required.
3. Generating synthetic data::
   • Select the "Synthetic Data Generation" function module.
   • Create and edit training data using interactive visualization tools.
   • Save the generated dataset for subsequent fine-tuning.
4. Teamwork::
   • Select the Dataset Collaboration functional module.
   • Use Git version control to manage datasets and facilitate team member collaboration.
   • Provide examples, tips, feedback, and other information about the dataset to make it easier for team members to work on it together.
5. Tip Generation::
   • Select the "Prompt Generation" function module.
   • Upload the dataset and select the type of prompt (e.g., chain thinking, fewer samples, multiple samples).
   • Automatic generation of hints for model training and inference.
6. Integration into Workflow::
   • Integrate Kiln's datasets and functionality into your own workflows using Kiln's Python library and the OpenAPI REST API.
   • Refer to Kiln's documentation and sample code to get started with development quickly.

Detailed Operation Procedure

• fine-tuned model: Details on how to select a model, upload data, configure parameters and initiate fine-tuning.
• Synthetic data generation: Details on how to create and edit data using visualization tools.
• Teamwork: A detailed description of how to use Git version control to manage datasets and how to provide and process feedback.
• Tip Generation: Details on how to select a prompt type, upload data, and generate a prompt.
• Integration into Workflow: Details how to use Python libraries and APIs for integration, providing sample code and usage scenarios.