Batched and differentiable inference

Looking for usage documentation? Check out Fine tuning.

`TabPFNRegressor.fit_from_preprocessed`

Used in Fine-Tuning. Fit the model to preprocessed inputs from torch dataloader inside a training loop a Dataset provided by get_preprocessed_datasets. This function always uses the “batched” fit_mode.

TabPFNRegressor.fit_from_preprocessed(
    X_preprocessed: list[torch.Tensor],
    y_preprocessed: list[torch.Tensor],
    cat_ix: list[list[list[int]]],
    configs: list[list[EnsembleConfig]],
    *,
    performance_options: PerformanceOptions,
    no_refit: bool = True,
) -> TabPFNRegressor

Parameters

Parameter	Type	Default	Description
`X_preprocessed`	`list[torch.Tensor]`	Required	The input features obtained from the preprocessed Dataset The list contains one item for each ensemble predictor. use tabpfn.utils.collate_for_tabpfn_dataset to use this function with batch sizes of more than one dataset (see examples/tabpfn_finetune.py)
`y_preprocessed`	`list[torch.Tensor]`	Required	The target variable obtained from the preprocessed Dataset
`cat_ix`	`list[list[list[int]]]`	Required	categorical indices obtained from the preprocessed Dataset
`configs`	`list[list[EnsembleConfig]]`	Required	Ensemble configurations obtained from the preprocessed Dataset
`performance_options`	`PerformanceOptions`	Required	Performance and memory options forwarded to the model on each forward call inside the resulting executor.
`no_refit`	`bool`	`True`	if `True`, the classifier will not be reinitialized when calling fit multiple times.

Returns

Type	Description
`TabPFNRegressor`	—

`TabPFNRegressor.fit_with_differentiable_input`

View source

Fit the model with differentiable input. Mirror of TabPFNClassifier.fit_with_differentiable_input. Lets gradients flow from a downstream loss back through X (and y, if it carries grads) into upstream torch modules. Use this instead of fit when differentiable_input=True.

TabPFNRegressor.fit_with_differentiable_input(
    X: torch.Tensor,
    y: torch.Tensor,
) -> Self

Parameters

Parameter	Type	Default	Description
`X`	`torch.Tensor`	Required	The input data as a torch tensor.
`y`	`torch.Tensor`	Required	The target variable as a torch tensor.

Returns

Type	Description
`Self`	self

`TabPFNRegressor.predict_batched`

View source

Predict for several independent datasets in one pass. Each triple is preprocessed exactly as fit() + predict() does, then compatible model-input shapes are fused. Heterogeneous post-preprocessing shapes run in separate groups. This is equivalent to independent prediction and leaves self unchanged. Datasets need not share a target scale (each is decoded with its own bar distribution) but must share raw array shapes. Fitted transforms may produce different shapes; these are grouped internally without padding.

TabPFNRegressor.predict_batched(
    X_train_list: list[XType],
    y_train_list: list[YType],
    X_test_list: list[XType],
    *,
    output_type: OutputType = "mean",
    quantiles: list[float] | None = None,
) -> list[RegressionResultType]

Type aliases

OutputType = Literal["mean", "median", "mode", "quantiles", "full", "main"]

Parameters

Parameter	Type	Default	Description
`X_train_list`	`list[XType]`	Required	Training features, one array per dataset (all same shape).
`y_train_list`	`list[YType]`	Required	Training targets, one array per dataset.
`X_test_list`	`list[XType]`	Required	Test features, one array per dataset (all same shape).
`output_type`	`OutputType`	`"mean"`	As in `predict`, applied to every dataset.
`quantiles`	`list[float] \| None`	`None`	As in `predict`.

Returns

Type	Description
`list[RegressionResultType]`	One entry per dataset, in input order, each being what `predict` would return for that dataset.

Raises ValueError If the input lists have unequal or zero length, or the training (or test) arrays do not all share one shape. TabPFNValidationError If output_type or quantiles are invalid. NotImplementedError If tuning_config is configured on the estimator — the calibrated ensemble temperature is per-dataset state and cannot be applied correctly across a shared batch, so score those datasets individually with predict. Also raised for inference_precision=torch.float64, which the fused forward does not support. Note Constant-target datasets are answered analytically and take no part in the fused forward.

`TabPFNRegressor.forward`

View source

Forward pass for TabPFNRegressor Inference Engine. Used in fine-tuning and prediction. Called directly in FineTuning training loop or by predict() function with the use_inference_mode flag explicitly set to True. Iterates over outputs of InferenceEngine.

TabPFNRegressor.forward(
    X: list[torch.Tensor] | XType,
    *,
    use_inference_mode: bool = False,
) -> tuple[torch.Tensor | None, list[torch.Tensor], list[np.ndarray]]

Parameters

Parameter	Type	Default	Description
`X`	`list[torch.Tensor] \| XType`	Required	list[torch.Tensor] in fine-tuning, `XType` in normal predictions.
`use_inference_mode`	`bool`	`False`	Flag for inference mode., default at `False` since

Returns

Type	Description
`tuple[torch.Tensor \| None, list[torch.Tensor], list[np.ndarray]]`	A tuple containing: - Averaged logits over the ensemble (for fine-tuning). - Raw outputs from each estimator in the ensemble. - Borders used for each estimator.

Reference

tabpfn

tabpfn-client

tabpfn-extensions

REST API

Batched and differentiable inference

`TabPFNRegressor.fit_from_preprocessed`

`TabPFNRegressor.fit_with_differentiable_input`

`TabPFNRegressor.predict_batched`

`TabPFNRegressor.forward`

​TabPFNRegressor.fit_from_preprocessed

​TabPFNRegressor.fit_with_differentiable_input

​TabPFNRegressor.predict_batched

​TabPFNRegressor.forward

`TabPFNRegressor.fit_from_preprocessed`

`TabPFNRegressor.fit_with_differentiable_input`

`TabPFNRegressor.predict_batched`

`TabPFNRegressor.forward`