Data and application limit requirements
This page is a reference for data and application limit requirements for a standard Snorkel Flow installation. When creating uploading new datasets and creating new applications, please refer to the these class, dataset, and datapoint limits for the best performance.
Snorkel Flow enforces a per-file limit size of 100 MB via the user interface. If you are uploading datasets via the SDK, this limit does not apply.
| # Unique classes | Max total dataset size | Max datapoint size (i.e., single row) | GPU required? | |
|---|---|---|---|---|
| Text classification (multi-class) | 2 - 100 | 2 GB | 10 KB | Recommended |
| Text classification (multi-label) | 1 - 100 | 2 GB | 10 KB | Recommended |
| Text extraction (candidate-based) | 2 - 100 | 2 GB | 10 KB | Recommended |
| Text extraction (sequence tagging) | 1 - 25 | 250 MB | 10 KB | Required |
| PDF extraction | 2 - 20 | 1.6 GB | 1.6 MB | Recommended |
| Image classification | 1 - 10 | 20 GB | 500 KB | Required |
Reach out to your Snorkel representative for more information about instance sizing.