| --- |
| language: en |
| tags: |
| - summarization |
| - t5 |
| - cnn-dailymail |
| datasets: |
| - abisee/cnn_dailymail |
| metrics: |
| - rouge |
| pipeline_tag: summarization |
| --- |
| |
| # T5-small Fine-Tuned on CNN/DailyMail |
|
|
| This model generates abstractive summaries of English news articles. |
|
|
| ## Model Details |
| - **Base model:** t5-small |
| - **Dataset:** CNN/DailyMail v3.0.0 |
| - **Task:** Abstractive text summarization |
| - **Framework:** PyTorch + HuggingFace Transformers |
|
|
| ## Performance (ROUGE, on held-out test sample) |
| | Metric | Score | |
| |---|---| |
| | ROUGE-1 | 36.1 | |
| | ROUGE-2 | 15.36 | |
| | ROUGE-L | 26.07 | |
|
|
| ## How to Use |
|
|
| ⚠️ This model requires the `"summarize: "` prefix on every input, since it's a T5 model. |
|
|
| ```python |
| from transformers import pipeline |
| |
| summarizer = pipeline("summarization", model="samandar1105/text-summarizer") |
| text = "summarize: " + "YOUR LONG ARTICLE TEXT HERE" |
| result = summarizer(text, max_length=80, min_length=20, do_sample=False) |
| print(result[0]["summary_text"]) |
| ``` |
|
|
| ## Training Details |
| - Learning rate: 3e-5 |
| - Epochs: 4 |
| - Batch size: 8 |
| - Max input length: 512 tokens |
| - Max target length: 128 tokens |
|
|
