| --- |
| title: FelaTab playground |
| colorFrom: indigo |
| colorTo: green |
| sdk: gradio |
| sdk_version: 4.44.0 |
| app_file: app.py |
| pinned: false |
| license: other |
| --- |
| |
| # FelaTab playground |
|
|
| Paste a small labelled table and a row to predict. FelaTab learns the pattern from your example |
| rows in a single forward pass, with no training or setup, and fills the answer with a |
| confidence range. Runs on CPU. |
|
|
| ## How to use |
|
|
| Put your data as comma or space separated rows, one row per line, with a header line. The last |
| column is the label. Use `?` in the label column for the rows you want predicted. |
|
|
| - If the label column holds categories (e.g. Apple / Lemon / Grape), it predicts a class with |
| probabilities (classification). |
| - If the label column holds numbers, it predicts a value with an error bar (regression). |
|
|
| ## Which model |
|
|
| The Space loads the small tier by default (dim512, about 51.6M parameters, int8, roughly 52 MB). |
| Set the environment variable `FELATAB_TIER=big` to load the flagship (about 411.9M parameters). |
|
|
| ## Honest scope |
|
|
| This is a research preview. FelaTab matches or slightly trails a tuned gradient boosted tree on |
| zero shot classification and is behind trees on regression accuracy (it ships calibrated error |
| bars for regression rather than headline accuracy). See the model card for the measured numbers. |
|
|
