Hugging Face Introduces AI Sheets, a No-Code Tool for Dataset Creation and Transformation
Created on August 20|Last edited on August 20
Comment
Hugging Face has released AI Sheets, an open-source no-code tool that makes it easier to build, transform, and enrich datasets using AI models. The interface is similar to a spreadsheet, allowing users to experiment quickly by generating new columns with simple prompts. This makes it possible to test models, refine prompts, clean data, classify text, or generate synthetic datasets without writing code. AI Sheets can be run locally or directly from the Hugging Face Hub.
How it Works
Users can either import an existing dataset in formats such as CSV, TSV, XLS, or Parquet, or generate one from scratch by describing it in natural language. Once data is loaded, new columns can be added with prompts that instruct the model to perform tasks such as summarization, translation, or categorization. AI-generated cells can be refined by editing, validating, or liking outputs, which then become examples that guide future generations. Users can also drag down cells to instantly expand datasets and toggle web search for up-to-date information.
Use Cases
AI Sheets has several practical applications. Researchers can compare models by running them on the same dataset and even add an LLM-as-judge column to evaluate responses. Businesses can improve prompts for customer support systems by iterating in real-time on real requests. Data scientists can clean noisy datasets, classify content, or extract key insights. The tool also enables enrichment, such as filling missing information, and can generate synthetic datasets for cases where real data is limited or sensitive.
Export and Scaling
Once satisfied with the dataset, users can export it to the Hugging Face Hub. This creates a reusable configuration file that can be applied to larger-scale generation pipelines via Hugging Face Jobs. This approach allows experimentation at small scale before scaling up to thousands of rows.
Examples in Action
Hugging Face has shared examples of AI Sheets in use, such as comparing open-source models for generating mini web apps and categorizing Hub datasets by topic. Other experiments show evaluation setups where one model acts as a judge of others. The flexibility of the spreadsheet-style interface combined with integration of thousands of models from the Hub makes it possible to tailor datasets for diverse projects.
Anyone can try AI Sheets for free through a demo space on Hugging Face or install it locally from GitHub. For more intensive use, the Hugging Face Pro plan offers increased inference capacity. With its balance of accessibility and depth, AI Sheets aims to lower the barrier to dataset creation and transformation, bringing AI-powered data workflows to a wider audience.
Add a comment
Tags: ML News
Iterate on AI agents and models faster. Try Weights & Biases today.