---
license: mit
tags:
- toxicity-detection
- perspective
- gemini-3.5-flash
pipeline_tag: text-classification
language:
- en
library_name: pytorch
---

# Perspective API - GEMINI-3.5-FLASH - Regression

Toxicity prediction model trained on the GEMINI-3.5-FLASH dataset.

| Property | Value |
|----------|-------|
| Model | Perspective API |
| Task | Regression |
| Dataset | gemini-3.5-flash |
| Framework | PyTorch / PyTorch Lightning |


## Model Information

See the ToxicThesis repository for model class documentation.



## Usage

```python
from huggingface_hub import hf_hub_download
import torch

checkpoint_path = hf_hub_download(
    repo_id="simocorbo/toxicthesis-gemini-3.5-flash-perspective-regression",
    filename="checkpoints/best.pt"
)

checkpoint = torch.load(checkpoint_path, map_location='cpu', weights_only=False)
print("Checkpoint keys:", checkpoint.keys())

# See ToxicThesis repository for model implementation
# git clone https://github.com/simo-corbo/ToxicThesis
```



## Score Interpretation

| Output | Range | Meaning |
|--------|-------|---------|
| `score` | [0, 1] | Continuous toxicity score. Higher = more toxic. |

**Thresholding**: Use `score >= 0.5` for binary toxic/non-toxic decision.


## Files

| File | Description |
|------|-------------|
| `checkpoints/best.pt` | Model checkpoint (best validation loss) |
| `hparams.yaml` | Hyperparameters used for training |
| `train.csv` | Training metrics per epoch |
| `val.csv` | Validation metrics per epoch |
| `vocab_stanza_hybrid.pkl` | Vocabulary (for tree-based models) |

## Installation

```bash
# Clone ToxicThesis for full model implementations
git clone https://github.com/simo-corbo/ToxicThesis
cd ToxicThesis
pip install -r requirements.txt

# Or install dependencies directly
pip install torch transformers huggingface_hub fasttext-wheel stanza
```

## Citation

```bibtex
@software{toxicthesis2025,
  title={ToxicThesis},
  author={Corbo, Simone},
  year={2025},
  url={https://github.com/simo-corbo/ToxicThesis}
}
```