> ## 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.

# Persistence and devices

> Save a fitted classifier, light wrapper around save_fitted_tabpfn_model.

> [!NOTE]
>
> Looking for usage documentation? Check out [Classification](https://docs.priorlabs.ai/capabilities/classification.md).

## `TabPFNClassifier.save_fit_state`

[ View source ](https://github.com/PriorLabs/TabPFN/blob/f02b3b5941717472eafd078c15ae2043676b3fcf/src/tabpfn/classifier.py#L2114)

Save a fitted classifier, light wrapper around [`save_fitted_tabpfn_model`](https://docs.priorlabs.ai/api-reference/python/tabpfn/model-persistence#save-fitted-tabpfn-model).

```python theme={null}
TabPFNClassifier.save_fit_state(
    path: Path | str,
) -> None
```

**Parameters**

| Parameter | Type | Default | Description |
| - | - | - | - |
| `path` | `Path \| str` | Required | — |

**Returns**

| Type | Description |
| - | - |
| `None` | — |

***

## `TabPFNClassifier.load_from_fit_state`

[ View source ](https://github.com/PriorLabs/TabPFN/blob/f02b3b5941717472eafd078c15ae2043676b3fcf/src/tabpfn/classifier.py#L2118)

Restore a fitted clf, light wrapper around [`load_fitted_tabpfn_model`](https://docs.priorlabs.ai/api-reference/python/tabpfn/model-persistence#load-fitted-tabpfn-model).

```python theme={null}
TabPFNClassifier.load_from_fit_state(
    path: Path | str,
    *,
    device: DevicesSpecification = "auto",
) -> TabPFNClassifier
```

> [accordion] Type aliases
>
> ```python theme={null}
> DevicesSpecification = torch.device | str | Sequence[torch.device | str] | Literal["auto"]
> ```

**Parameters**

| Parameter | Type | Default | Description |
| - | - | - | - |
| `path` | `Path \| str` | Required | — |
| `device` | `DevicesSpecification` | `"auto"` | — |

**Returns**

| Type | Description |
| - | - |
| `TabPFNClassifier` | — |

***

## `TabPFNClassifier.to`

[ View source ](https://github.com/PriorLabs/TabPFN/blob/f02b3b5941717472eafd078c15ae2043676b3fcf/src/tabpfn/classifier.py#L2130)

Move the estimator to the given device(s).

If "auto": devices are selected based on availability in the
following order of priority: all available CUDA GPUs, "mps", "cpu".

To manually select a single device: specify a PyTorch device string e.g.
"cuda:1". See PyTorch's documentation for information about supported
devices.

To use several GPUs: specify a list of PyTorch GPU device strings, e.g.
\["cuda:0", "cuda:1"]. This can dramatically speed up inference for
larger datasets, by executing the estimators in parallel on the GPUs.
Multiple GPUs are only used when `fit_mode="fit_preprocessors"` or
`fit_mode="low_memory"`. In other cases, only the first GPU is used.

```python theme={null}
TabPFNClassifier.to(
    device: DevicesSpecification,
) -> None
```

> [accordion] Type aliases
>
> ```python theme={null}
> DevicesSpecification = torch.device | str | Sequence[torch.device | str] | Literal["auto"]
> ```

**Parameters**

| Parameter | Type | Default | Description |
| - | - | - | - |
| `device` | `DevicesSpecification` | Required | — |

**Note**

The specified device is only used once the model is initialized. This occurs
during the first .fit() call.

**Returns**

| Type | Description |
| - | - |
| `None` | — |


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.