(CHANGE BASE AND REBASE) Include server error forwarding

[liam-sbhoo]

Development

With the development of server (PR 1 and PR 2)

• The client can now raise inference-related errors (raised on the server) to the users

Question

• Calling fit with conflicting X_train and y_train wouldn't raise an error until predict is called - calling fit would only upload the dataset, but not actually fitting it on TabPFN.
• Should we change this?

Example

The following demonstrates the behavior when the client calls predict with conflicting dataset:

# Load data
from sklearn.datasets import load_breast_cancer, load_digits
from sklearn.model_selection import train_test_split

breast_X, breast_y = load_breast_cancer(return_X_y=True)
breast_X_train, breast_X_test, breast_y_train, breast_y_test = train_test_split(X, y)
digits_X, digits_y = load_digits(return_X_y=True)
digits_X_train, digits_X_test, digits_y_train, digits_y_test = train_test_split(digits_X, digits_y)

# Init TabPFN
from tabpfn_client import tabpfn_classifier
tabpfn_classifier.init(use_server=True)

# Predict with TabPFN
from tabpfn_client import TabPFNClassifier

tabpfn = TabPFNClassifier()
tabpfn.fit(breast_X_train, breast_y_train)
tabpfn.predict(digits_X_test)

Outcome at the client:

[liam-sbhoo]

• Change PR base and rebase
• Update submodule upon merging common

[liam-sbhoo]

Not relevant anymore.