Connect a custom model

Image classification

Custom PyTorch image classifiers are currently documented. Supply the model, its matching preprocessing pipeline, and optional category labels.

Follow the custom PyTorch model guide

LLM providers

The LLM explainer accepts built-in provider identifiers and provider-specific client arguments. See provider configuration.

Image generation and editing

ImageGenerationAndEditingExplainer supports multiple integration routes, including supported providers, a pre-loaded compatible pipeline, or a custom model/generation function.

For a custom integration, provide a model or pipeline together with a compatible custom_generate_fn that XWhy can call when producing the reference and perturbed outputs. The explainer then applies its perturbation, output-distance, and surrogate-modelling workflow around that interface.

See the image generation and editing guide.

Tabular models

TabularExplainer is available for structured classification and regression. Pass the trained black-box model to the explainer and ensure its prediction interface is compatible with the XWhy tabular adapter. Use the same preprocessing and feature scaling used by the model during normal inference.

See the tabular explainer guide.

Text models

TextExplainer accepts either a model exposing predict_proba, predict, or __call__, or a direct predict_fn that accepts a sequence of texts and returns predictions or scores.

See the text explainer guide.

Development and roadmap modalities

Not yet supported end to end

PointCloudExplainer remains a development interface. Time Series and Multimodal explainability are planned capabilities. Agentic AI and Multi-Agent AI are currently research-roadmap areas rather than exported explainers.