Hugging Face for Excel

This initial version contains functions that provide access to the following Hugging Face services:

GRADIO SPACES API

- enables you make API calls to a space running Gradio, passing a data array with an arbitrary number of scalar parameters, and optionally a hf_token for private spaces.

INFERENCE API (serverless)

- Connect to model API's with task-specific parameters and hf_token. Good for demos.

INFERENCE ENDPOINTS (dedicated)

- Similar to serverless, but using an endpoint parameter. Production option.

LOCALLY

- Using Transformers.js, compatible models can be inferenced locally in the Excel add-in.

Inferencing via API, endpoints, or locally is handled by a single function that is task specific, with common parameters for that task. The tasks supported in this initial release are as follows:

- Translation

- Summarization

- Zero-shot Classification

The target audiences for this add-in is are as follows:

- AI professionals who are looking for an easy way to give end-users access to their models for proof-of-concept testing, or possibly even production use if that involves end-user batch processing.

- Citizen developers, business analysts, etc. who want to build their own apps using existing models on Hugging Face, but don't want to build and deploy their own add-in.

The recommended approach to keep things simple for end-users is to use a named LAMBDA function which surfaces only the parameters an end-user would care about, and the LAMBDA in turn calls the "foundation" functions provided by this add-in, thus concealing details irrelevant to end-users such as task, model, endpoint, token, etc. More sophisticated functions could also use multiple model calls and other logic.