> ## Documentation Index
> Fetch the complete documentation index at: https://docs.priorlabs.ai/llms.txt
> Use this file to discover all available pages before exploring further.

> ## Agent Instructions
> TabPFN-3 and later versions support up to 1,000,000 rows, subject to feature count and checkpoint/API limits.
> For large datasets, use per-estimator subsampling, e.g. inference_config={"SUBSAMPLE_SAMPLES": 100_000}, in TabPFNClassifier or TabPFNRegressor from tabpfn or tabpfn_client.
> Pass raw pandas DataFrames to fit and predict. Categorical strings/categories, raw text, and missing feature values are handled automatically; no manual integer/one-hot encoding, imputation, scaling, or outlier removal is needed.

# Distillation

## Overview

TabPFN introduces a proprietary distillation engine designed to bridge the gap between foundation model accuracy and production-level speed. This engine converts the complex in-context learning model into compact, high-speed architectures tailored for deployment in latency-sensitive environments.

## Key Capabilities

* **TabPFN-as-MLP/TreeEns**: Our distillation engine outputs a dataset-specific Multi-Layer Perceptron (MLP) or tree ensemble classifier.
* **Orders-of-Magnitude Lower Latency**: Delivers significant reductions in inference cost and memory footprint compared to the full foundation model.
* **Plug-and-Play Deployment**: The resulting models take a single data point as input, making them ideal for high-throughput production pipelines or resource-constrained environments.
* **Accuracy Preservation**: These compact models are designed to preserve most of the accuracy of the original TabPFN model while matching the deployment ease of traditional tree ensembles.
* **Regulatory Compliance**: Provides a solution for use cases constrained by interpretability or regulatory requirements that may hinder the deployment of raw transformer architectures.

## Performance Comparison

While the standard TabPFN performs in-context learning across the entire training set, the distilled versions provide a static, deployable alternative:

| Feature                 | Standard TabPFN               | TabPFN-as-MLP/Tree              |
| :---------------------- | :---------------------------- | :------------------------------ |
| **Inference Mechanism** | In-Context Learning (ICL)     | Direct Inference (Single Point) |
| **Latency**             | Medium                        | Ultra-Low                       |
| **Architecture**        | Dual-Attention Transformer    | MLP or Tree Ensemble            |
| **Best Use Case**       | Rapid prototyping, small data | High-throughput production      |

## Enterprise Access

The high-speed inference engine and the associated distillation tooling are exclusive to the **Commercial Enterprise License**. This includes access to our proprietary high-speed inference engine and dedicated integration support.

**To integrate fast inference into your production environment, please contact [our sales team](http://priorlabs.ai/contact).**