---
title: Transformer
emoji: 🌍
colorFrom: pink
colorTo: green
sdk: docker
pinned: false
license: mit
---

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

To force a specific transformer model in Spaces, set the `MODEL_NAME` environment variable, for example:

`fnlp/bart-base-chinese`

## Training and evaluation

For local fine-tuning and metric collection:

```bash
python train.py --train-path data/train.jsonl --valid-path data/valid.jsonl --output-dir outputs/bart_cn
python evaluate.py --test-path data/test.jsonl --model-name outputs/bart_cn --output-csv metrics_report.csv
```

The evaluation script prints and exports:

- `ROUGE-L`
- `BERTScore`
- `QAFactEval` when an external QAFactEval environment is available
- length hit rate
- average latency