|
--- |
|
tags: |
|
- generated_from_trainer |
|
datasets: |
|
- cnn_dailymail |
|
- xsum |
|
- samsum |
|
- billsum |
|
- lytang/MeetingBank-transcript |
|
metrics: |
|
- rouge |
|
model-index: |
|
- name: t5_xsum_samsum_billsum_cnn_dailymail |
|
results: |
|
- task: |
|
name: Sequence-to-sequence Language Modeling |
|
type: text2text-generation |
|
dataset: |
|
name: cnn_dailymail |
|
type: cnn_dailymail |
|
config: 3.0.0 |
|
split: train |
|
args: 3.0.0 |
|
metrics: |
|
- name: Rouge1 |
|
type: rouge |
|
value: 0.2373 |
|
license: mit |
|
language: |
|
- en |
|
library_name: transformers |
|
pipeline_tag: summarization |
|
--- |
|
|
|
# t5_xsum_samsum_billsum_cnn_dailymail |
|
|
|
The `t5_xsum_samsum_billsum_cnn_dailymail` model is a text summarization model fine-tuned on the `t5-base` architecture, which is a versatile text-to-text transfer transformer. This powerful model excels at generating abstractive summaries from input text. It has been fine-tuned on multiple datasets, including CNN/Daily Mail (cnn_dailymail), XSum (xsum), SamSum (samsum), BillSum (billsum), and the MeetingBank-transcript dataset by lytang. |
|
|
|
## Intended Uses & Limitations |
|
|
|
### Intended Uses |
|
|
|
- Document summarization: The model is well-suited for summarizing lengthy documents or articles, making it valuable for content curation and information extraction tasks. |
|
- Content generation: It can be used to generate concise summaries from input text, which is useful for creating short and informative snippets. |
|
|
|
### Limitations |
|
|
|
- Model size: The model's size may require significant computational resources for deployment, limiting its use in resource-constrained environments. |
|
- Domain-specific content: While it performs well on general text summarization tasks, its performance may vary when applied to domain-specific content. |
|
|
|
## Training and Evaluation Data |
|
|
|
The model has been trained on a diverse set of datasets, including CNN/Daily Mail, XSum, SamSum, BillSum, and the MeetingBank-transcript dataset. These datasets provide a wide range of text summarization examples, enabling the model to generalize across various domains and styles of text. |
|
|
|
### Training hyperparameters |
|
|
|
The following hyperparameters were used during training: |
|
- learning_rate: 2e-05 |
|
- train_batch_size: 2 |
|
- eval_batch_size: 2 |
|
- seed: 42 |
|
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 |
|
- lr_scheduler_type: linear |
|
- num_epochs: 1 |
|
|
|
### Training results |
|
|
|
#### samsum |
|
|
|
| Rouge1 | Rouge2 | RougeL | RougeLsum | |
|
|:-------:|:-------:|:-------:|:---------:| |
|
| 0.0138 | 0.0002 | 0.0138 | 0.0138 | |
|
|
|
|
|
#### CNN_Dailymail |
|
|
|
| Training Loss | Epoch | Step | Validation Loss | Rouge1 | Rouge2 | Rougel | Rougelsum | Gen Len | |
|
|:-------------:|:-----:|:-----:|:---------------:|:------:|:------:|:------:|:---------:|:-------:| |
|
| 1.8486 | 1.0 | 32300 | 1.6478 | 0.2373 | 0.1086 | 0.1972 | 0.1971 | 18.9674 | |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.33.0 |
|
- Pytorch 2.0.0 |
|
- Datasets 2.1.0 |
|
- Tokenizers 0.13.3 |