README.md

---
language: en
license: apache-2.0
tags:
  - pytorch
  - text-classification
  - nli
  - zero-shot-classification
  - dei
  - equibert
metrics:
  - f1
  - accuracy
---

# EquiBERT — DEI Natural Language Inference

**Model ID:** `SallySims/equibert-nli`

DEI-domain NLI model for textual entailment and zero-shot
classification. Drop-in replacement for `facebook/bart-large-mnli`.

## Labels

| ID | Label | Description |
|----|-------|-------------|
| 0 | `entailment` | Premise logically supports hypothesis |
| 1 | `contradiction` | Premise contradicts hypothesis |
| 2 | `neutral` | Premise neither supports nor contradicts |

## Usage — Zero-Shot Classification

```python
from transformers import pipeline

nli = pipeline("zero-shot-classification", model="SallySims/equibert-nli")

result = nli(
    "We conduct annual pay equity reviews published in our DEI report.",
    candidate_labels=["pay equity", "hiring bias", "inclusion culture"]
)
# {"labels": ["pay equity", ...], "scores": [0.91, ...]}
```

## Usage — Direct Entailment

```python
premise    = "We conduct annual pay equity reviews."
hypothesis = "The organisation has a formal pay equity process."
inputs = tokenizer(premise, hypothesis, return_tensors="pt")
# label = id2label[model(**inputs).logits.argmax(-1).item()]
# → "entailment"
```

## DEI Policy Verification

Use this model to verify whether organisational statements are
consistent with stated DEI policies — identifying contradictions
between policy documents and actual communications.

## Model Description

EquiBERT is a multi-task DEI (Diversity, Equity and Inclusion) transformer
built on a dual-encoder backbone that fuses **RoBERTa-base** and
**DeBERTa-v3-base** via a learned weighted sum (α parameter).
The fused representation is fed into task-specific heads covering
17 distinct DEI analysis tasks.

**Organisation:** [SallySims](https://huggingface.co/SallySims)
**Framework:** PyTorch + HuggingFace Transformers
**Backbone:** RoBERTa-base + DeBERTa-v3-base (dual encoder, fused)
**Language:** English
**Domain:** Organisational DEI text — HR communications, policies,
job descriptions, performance reviews, leadership statements, reports

## Architecture

```
Input Text
    │
    ├──▶ RoBERTa-base encoder ──▶ Linear projection
    │                                     │
    └──▶ DeBERTa-v3-base encoder ──▶ Linear projection
                                          │
                              Weighted fusion (learned α)
                                          │
                                   Layer Norm + Dropout
                                          │
                              Task-specific head (see below)
```

## Training Data

Trained on synthetic DEI organisational text generated by the
EquiBERT synthetic data pipeline, covering 20 DEI categories
across HR, policy, leadership, and workforce analytics domains.
For production use, fine-tune on real labelled DEI data.

## Limitations

- Trained on synthetic data — predictions should be validated
  before use in real HR or policy decisions.
- English-only.
- Not a substitute for qualified DEI practitioners or legal advice.
- May reflect biases present in the training corpus.

## Citation

If you use EquiBERT in your research, please cite:

```bibtex
@misc{equibert2024,
  author    = {SallySims},
  title     = {EquiBERT: A Multi-Task DEI Transformer},
  year      = {2024},
  publisher = {HuggingFace},
  url       = {https://huggingface.co/SallySims}
}
```

## All EquiBERT Models

| Model | Task | Primary Metric |
|-------|------|---------------|
| [equibert-bias-classifier](https://huggingface.co/SallySims/equibert-bias-classifier) | Bias Detection | Macro F1 |
| [equibert-microaggression](https://huggingface.co/SallySims/equibert-microaggression) | Microaggression Detection | Macro F1 |
| [equibert-category-tagger](https://huggingface.co/SallySims/equibert-category-tagger) | DEI Category Tagging | Macro F1 |
| [equibert-event-exclusion](https://huggingface.co/SallySims/equibert-event-exclusion) | Event Exclusion Classification | Macro F1 |
| [equibert-inclusive-language](https://huggingface.co/SallySims/equibert-inclusive-language) | Inclusive Language Scoring | Span F1 |
| [equibert-review-auditor](https://huggingface.co/SallySims/equibert-review-auditor) | Performance Review Auditing | Span F1 |
| [equibert-washing-detector](https://huggingface.co/SallySims/equibert-washing-detector) | DEI Washing Detection | MAE |
| [equibert-framing-scorer](https://huggingface.co/SallySims/equibert-framing-scorer) | Report Framing Scoring | MAE |
| [equibert-awareness-scorer](https://huggingface.co/SallySims/equibert-awareness-scorer) | DEI Awareness Scoring | MAE |
| [equibert-similarity](https://huggingface.co/SallySims/equibert-similarity) | Semantic Similarity | Accuracy |
| [equibert-ner](https://huggingface.co/SallySims/equibert-ner) | DEI Entity Recognition | Span F1 |
| [equibert-relation-extraction](https://huggingface.co/SallySims/equibert-relation-extraction) | Relation Extraction | Macro F1 |
| [equibert-qa](https://huggingface.co/SallySims/equibert-qa) | Extractive QA | Span EM |
| [equibert-search](https://huggingface.co/SallySims/equibert-search) | Semantic Search | MRR@10 |
| [equibert-nli](https://huggingface.co/SallySims/equibert-nli) | NLI / Textual Entailment | Macro F1 |
| [equibert-generator](https://huggingface.co/SallySims/equibert-generator) | DEI Text Generation | ROUGE-L |