FastDatasets
FreeA powerful tool for creating high-quality training datasets for Large Language Models
About FastDatasets
FastDatasets is an open-source tool under Apache 2.0 that streamlines the creation of high-quality training datasets for fine-tuning Large Language Models (LLMs). It ingests documents in multiple formats (PDF, Word, Markdown, plain text), automatically splits them into chunks, and uses an LLM to generate relevant questions and answers. The tool outputs datasets ready for supervised fine-tuning (SFT) in popular formats like Alpaca and ShareGPT. It also supports knowledge distillation, allowing users to extract knowledge from larger models or augment existing datasets with instruction variants. A built-in web UI provides a no-code interface for document uploads, parameter configuration, real-time progress monitoring, and result downloads.
Key Features
Pros & Cons
- Open-source with permissive Apache 2.0 license, fully free to use.
- Supports common document formats (PDF, Word, Markdown, text) without conversion.
- Produces datasets directly in Alpaca and ShareGPT formats, compatible with many LLM fine-tuning frameworks.
- Includes both a web UI for casual use and CLI scripts for automation.
- Provides knowledge distillation and data augmentation features to enhance dataset quality and diversity.
- Asynchronous processing speeds up large-scale dataset generation.
- Requires external LLM API (e.g., DeepSeek) to be configured; no built-in language model.
- Quality of generated datasets heavily depends on the underlying LLM's capabilities.
- No native training or evaluation tools — only dataset creation.
- May require significant memory and time for very large documents or high concurrency.
- Documentation primarily in Chinese; English README available but less comprehensive.