Skip to main content

​
Model

An LLM predicts the next word while a Tabular Foundation Model (TFM) predicts the next value. Due to this, LLMs lose the structure that makes tabular data meaningful. Because Hollerith operates on rows and columns rather than flattening them into a prompt, it can capture the statistical relationships between fields that language models discard.That means real prediction as a true regression or classification task with high accuracy, with no dataset specific training required.

​
Getting set up

No. Inference runs solely on Monarcha’s infrastructure.
Python 3.11 or 3.12. See the Python SDK reference.

​
Your data

No. Your table is read as context during inference, but data is purged afterward.
Both are accepted, and both are ordinal-encoded. The model sees which rows share a value, not what the value means or how the values order.
No. String columns are read as labels; missing values are read as missing, which is information. An imputed median tells the model a value was observed when it was not.

​
Input size

There is no minimum. fit needs one labeled row and one feature column beside the target; evaluate needs two rows to split.
1 000 000 training rows, 2 000 columns and 100 000 000 cells; a table sits under all three, and the cell budget usually binds first — at 2 000 columns you get 50 000 rows, not 1 000 000. predict scores at most 200 000 rows per call, so a million rows to score is five calls. See Limits for more.

​
API calls

Yes, the same rows scored against the same fitted context will return the same predictions.
Seconds to minutes, with latency scaling as the amount of input data increases.
Authenticated API requests are limited to 60 per minute per user within an organization. See Limits.

​
Next steps

Quickstart

Make your first API call to Hollerith

Troubleshooting

Symptom-to-fix tables for common issues