README.md

---
title: Agentic NLI
emoji: 🔗
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 4.44.0
app_file: app.py
pinned: false
---

# AFVE NLI Space

This folder is the source-of-truth for the Hugging Face Space that serves AFVE NLI scoring over HTTP.

The Space mirrors the local AFVE NLI pipeline shape: it accepts a JSON list of `{ "text": premise, "text_pair": hypothesis }` objects and returns the same list of `{ "label": string, "score": number }` objects produced by the local Transformers `text-classification` pipeline.

The model defaults to `cross-encoder/nli-deberta-v3-base` to match the current backend AFVE setting. To run `cross-encoder/nli-deberta-v3-small` instead, set `NLI_MODEL_NAME=cross-encoder/nli-deberta-v3-small` in the Space environment.

Render switch-over:

- `NLI_BACKEND=remote`
- `NLI_SPACE_URL=<your Hugging Face Space URL>`

Local development remains unchanged when `NLI_BACKEND` is unset. To test the Space locally, set both variables in your local `.env`.

Space sync:

Fill in the actual Space repo URL and preferred sync command:

```bash
# Option A: copy this folder into the Space repo and push there.
# Option B: git subtree push --prefix hf-services/afve-nli <SPACE_GIT_REMOTE_URL> main
```