This Model is a fine-tuned version of T-systems summarization model v1.
We used 1000 examples of headline-content pairs from BR24 articles for the fine-tuning process.
Despite the small amount of training data, the tonality of the summarizations has changed significantly. Many of the resulting summaries do sound like a headline.
Training
We used the following parameters for training this model:
- base model: deutsche-telekom/mt5-small-sum-de-en-v1
- source_prefix: "summarize: "
- batch size: 4
- max_source_length: 400
- max_target_length: 35
- weight_decay: 0.01
- number of train epochs: 1
- learning rate: 5e-5
License
Since the base model is trained on the MLSUM dataset, this model may not be used for commercial use.
Stats
| Model | Rouge1 | Rouge2 | RougeL | RougeLSum |
|---|---|---|---|---|
| headlines_test_small_example | 13.573500 | 3.694700 | 12.560600 | 12.60000 |
| deutsche-telekom/mt5-small-sum-de-en-v1 | 10.6488 | 2.9313 | 10.0527 | 10.0523 |