Spaces:
Running
Running
| <html lang="en"> | |
| <head> | |
| <meta charset="utf-8" /> | |
| <meta name="viewport" content="width=device-width, initial-scale=1" /> | |
| <title>Kumo Tabular in the Browser</title> | |
| <meta name="description" content="Zero-shot tabular predictions on your own CSV or Parquet file with NVIDIA Kumo Tabular, DuckDB-WASM and WebGPU. Nothing leaves your browser." /> | |
| <link rel="icon" href="data:image/svg+xml,%3Csvg xmlns=%22http://www.w3.org/2000/svg%22 viewBox=%220 0 16 16%22%3E%3Crect width=%2216%22 height=%2216%22 rx=%223%22 fill=%22%2376b900%22/%3E%3Cpath d=%22M4 4h8M4 8h8M4 12h8%22 stroke=%22white%22 stroke-width=%221.5%22/%3E%3C/svg%3E" /> | |
| <script type="module" crossorigin src="/assets/index-1X7aveGr.js"></script> | |
| <link rel="stylesheet" crossorigin href="/assets/index-BXYgCNdU.css"> | |
| </head> | |
| <body> | |
| <header class="top"> | |
| <div class="brand"> | |
| <h1>Kumo Tabular <span>in the browser</span></h1> | |
| <p> | |
| Drop a CSV or Parquet file, pick a column, get predictions. NVIDIA's | |
| <a href="https://huggingface.co/nvidia/Kumo-Tabular" target="_blank" rel="noreferrer">Kumo Tabular</a> | |
| learns from your labeled rows in one forward pass: no training, no server. Your data stays on this page. | |
| </p> | |
| </div> | |
| <div class="badges"> | |
| <span id="device-badge" class="badge">checking GPU…</span> | |
| <span class="badge">DuckDB-WASM</span> | |
| <span class="badge">Small · 28M</span> | |
| </div> | |
| </header> | |
| <main> | |
| <section class="card" id="data-card"> | |
| <div class="card-head"> | |
| <h2><span class="step">1</span> Data</h2> | |
| <span id="data-summary" class="muted"></span> | |
| </div> | |
| <label id="drop" class="drop"> | |
| <input id="file" type="file" accept=".csv,.parquet,.tsv,.txt" hidden /> | |
| <strong>Drop a CSV or Parquet file</strong> | |
| <span class="muted">or click to choose one. It is read locally by DuckDB.</span> | |
| </label> | |
| <div class="samples"> | |
| <span class="muted">Or try:</span> | |
| <button data-sample="churn_synthetic.csv" data-target="churn">Churn (150 unlabeled rows)</button> | |
| <button data-sample="breast_cancer.csv" data-target="diagnosis">Breast cancer</button> | |
| <button data-sample="iris.csv" data-target="species">Iris</button> | |
| <button data-sample="california_housing.csv" data-target="MedHouseVal">California housing</button> | |
| <button data-sample="diabetes.csv" data-target="progression">Diabetes</button> | |
| </div> | |
| <details id="sql-details"> | |
| <summary>Filter or reshape with SQL</summary> | |
| <p class="muted">The model sees the result of this query. The file is table <code>source</code>.</p> | |
| <textarea id="sql" spellcheck="false">SELECT * FROM source</textarea> | |
| <button id="apply-sql" class="secondary">Apply</button> | |
| </details> | |
| </section> | |
| <section class="card" id="setup-card" hidden> | |
| <div class="card-head"> | |
| <h2><span class="step">2</span> What to predict</h2> | |
| </div> | |
| <div class="setup"> | |
| <label>Target column <select id="target"></select></label> | |
| <label>Task | |
| <select id="task"> | |
| <option value="classification">Classification (up to 10 classes)</option> | |
| <option value="regression">Regression</option> | |
| </select> | |
| </label> | |
| <label>Rows to predict | |
| <select id="mode"> | |
| <option value="missing">Rows where the target is empty</option> | |
| <option value="holdout">Hold out 20% and score them</option> | |
| </select> | |
| </label> | |
| </div> | |
| <p id="split-note" class="muted"></p> | |
| <div class="columns-head"> | |
| <span>Features</span> | |
| <span class="muted">numbers are numerical; text and booleans are categorical</span> | |
| </div> | |
| <div id="columns" class="columns"></div> | |
| <div class="run-row"> | |
| <button id="run" class="primary">Predict</button> | |
| <div id="progress" class="progress" hidden><div></div></div> | |
| <span id="status" class="muted"></span> | |
| </div> | |
| </section> | |
| <section class="card" id="results-card" hidden> | |
| <div class="card-head"> | |
| <h2><span class="step">3</span> Predictions</h2> | |
| <span id="timing" class="muted"></span> | |
| </div> | |
| <div id="metrics" class="metrics"></div> | |
| <div class="results-grid"> | |
| <div id="chart" class="chart"></div> | |
| <div class="table-wrap"><table id="preview"></table></div> | |
| </div> | |
| <details open> | |
| <summary>Query the predictions with SQL</summary> | |
| <p class="muted">Tables: <code>source</code>, <code>data</code>, <code>predictions</code>, and <code>results</code> (data joined with predictions).</p> | |
| <textarea id="query" spellcheck="false"></textarea> | |
| <div class="run-row"> | |
| <button id="run-query" class="secondary">Run query</button> | |
| <button id="download" class="secondary">Download results.csv</button> | |
| </div> | |
| <div class="table-wrap"><table id="query-out"></table></div> | |
| </details> | |
| </section> | |
| </main> | |
| <footer> | |
| Model: <a href="https://huggingface.co/nvidia/Kumo-Tabular" target="_blank" rel="noreferrer">nvidia/Kumo-Tabular</a> | |
| (Small, OpenMDW-1.1), converted to ONNX. Preprocessing ports | |
| <a href="https://github.com/NVIDIA/structured-data-models" target="_blank" rel="noreferrer">structured-data-models</a>' | |
| default recipe (one estimator) and matches it value for value. Runs with | |
| <a href="https://huggingface.co/docs/transformers.js" target="_blank" rel="noreferrer">Transformers.js</a> and | |
| <a href="https://duckdb.org/docs/api/wasm/overview" target="_blank" rel="noreferrer">DuckDB-WASM</a>. | |
| </footer> | |
| </body> | |
| </html> | |