README.md

---
license: apache-2.0
base_model: phobert-v2
tags:
- vietnamese
- hate-speech
- span-detection
- token-classification
- nlp
datasets:
- visolex/ViHOS
model-index:
- name: phobert-v2-hsd-span
  results:
  - task:
      type: token-classification
      name: Hate Speech Span Detection
    dataset:
      name: visolex/ViHOS
      type: visolex/ViHOS
    metrics:
      - type: f1
        value: 0.6326
      - type: precision
        value: 0.6494
      - type: recall
        value: 0.6305
      - type: exact_match
        value: 0.0000
---

# phobert-v2-hsd-span: Hate Speech Span Detection (Vietnamese)

This model is a fine-tuned version of [phobert-v2](https://huggingface.co/phobert-v2) for Vietnamese **Hate Speech Span Detection**.

## Model Details

- Base Model: `phobert-v2`
- Description: Vietnamese Hate Speech Span Detection
- Framework: HuggingFace Transformers
- Task: Hate Speech Span Detection (token/char-level spans)

### Hyperparameters

- Max sequence length: `64`
- Learning rate: `5e-6`
- Batch size: `32`
- Epochs: `100`
- Early stopping patience: `5`

## Results

- F1: `0.6326`
- Precision: `0.6494`
- Recall: `0.6305`
- Exact Match: `0.0000`

## Usage

```python
from transformers import AutoTokenizer, AutoModelForTokenClassification
import torch

model_name = "phobert-v2-hsd-span"
tok = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)
text = "Ví dụ câu tiếng Việt có nội dung thù ghét ..."
enc = tok(text, return_tensors="pt", truncation=True, max_length=256, is_split_into_words=False)
with torch.no_grad():
    logits = model(**enc).logits
    pred_ids = logits.argmax(-1)[0].tolist()
# TODO: chuyển pred_ids -> spans theo scheme nhãn của bạn (BIO/BILOU/char-offset)
```

## License

Apache-2.0

## Acknowledgments

- Base model: [phobert-v2](https://huggingface.co/phobert-v2)