README.md

---
language:
  - en
license: mit
library_name: transformers
pipeline_tag: zero-shot-classification
tags:
  - zero-shot
  - multi-label
  - text-classification
  - pytorch
metrics:
  - precision
  - recall
  - f1
base_model: bert-base-uncased
datasets:
  - polodealvarado/zeroshot-classification
---

# Zero-Shot Text Classification — projection_biencoder

CLIP-inspired with projection heads, L2 norm, and learnable temperature.

This model encodes texts and candidate labels into a shared embedding space using BERT,
enabling classification into arbitrary categories without retraining for new labels.

## Training Details

| Parameter | Value |
|-----------|-------|
| Base model | `bert-base-uncased` |
| Model variant | `projection_biencoder` |
| Training steps | 1000 |
| Batch size | 2 |
| Learning rate | 2e-05 |
| Trainable params | 109,679,105 |
| Training time | 318.3s |

## Dataset

Trained on [polodealvarado/zeroshot-classification](https://huggingface.co/datasets/polodealvarado/zeroshot-classification).

## Evaluation Results

| Metric | Score |
|--------|-------|
| Precision | 0.9431 |
| Recall | 0.9826 |
| F1 Score | 0.9624 |

## Usage

```python
from models.projection import ProjectionBiEncoderModel

model = ProjectionBiEncoderModel.from_pretrained("polodealvarado/projection_biencoder")

predictions = model.predict(
    texts=["The stock market crashed yesterday."],
    labels=[["Finance", "Sports", "Biology", "Economy"]],
)
print(predictions)
# [{"text": "...", "scores": {"Finance": 0.98, "Economy": 0.85, ...}}]
```