Downloads · 30 days
11
9% of all-time downloads
tibetan-metadata-roberta-ner
ganga4364/tibetan-metadata-roberta-ner
tibetan-metadata-roberta-ner is a token classification model from ganga4364. Use it when you need labels on individual words, such as names. It is set up for transformers. The card lists the license as apache-2.0.
Fine-tuned spsither/tibetanRoBERTaSe3 for title and author span detection in Tibetan text segments.
All-time downloads
121
Public
Parameters
82.9M
663 MB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors331 MB · 98%
At a glance
- Task
- Token Classification
- Library
- transformers
- License
- apache-2.0
- Model type
- roberta
- Access
- Public
- Created
- Jun 12, 2026
- Updated
- Jun 16, 2026
- SHA
- 8d11b041
Base models
Model card
From the Hugging Face model README
Tibetan Metadata RoBERTa NER
Fine-tuned spsither/tibetan_RoBERTa_S_e3 for title and author span detection in Tibetan text segments.
Labels (BIO)
| ID | Label |
|---|---|
| 0 | O |
| 1 | B-TITLE |
| 2 | I-TITLE |
| 3 | B-AUTHOR |
| 4 | I-AUTHOR |
Training
- Dataset: ganga4364/tibetan-metadata-detector
- 3 epochs, batch size 64, lr 2e-5
- Sliding-window examples: 512 tokens, stride 256, 15B+15E overlap-aware
Test metrics (full test split, 108,954 windows)
| Metric | Value |
|---|---|
| span F1 | 14.9% |
| span precision | 11.9% |
| span recall | 20.0% |
| title F1 | 16.5% |
| author F1 | 1.4% |
Inference
Important: Do not pass entire long segments in one forward pass. Use the same sliding-window pipeline as training (15 begin + 15 end, stride 256), merge overlapping predictions, then extract spans.
from pipeline.inference import load_model_and_tokenizer, predict_segment, highlight_spans
model, tokenizer, device = load_model_and_tokenizer("ganga4364/tibetan-metadata-roberta-ner")
spans = predict_segment(model, tokenizer, your_tibetan_text, device=device)
print(highlight_spans(your_tibetan_text, spans))
Demo
Try the Gradio Space: ganga4364/tibetan-metadata-highlight
- Task
- Token Classification
- Library
- transformers
- Type
- roberta
- License
- apache-2.0
- Languages
- bo
- Created
- Jun 12, 2026
- Updated
- Jun 16, 2026