|
--- |
|
license: apache-2.0 |
|
tags: |
|
- summarization |
|
- generated_from_trainer |
|
metrics: |
|
- rouge |
|
model-index: |
|
- name: flan-t5-base-xsum-swe |
|
results: [] |
|
|
|
inference: |
|
parameters: |
|
temperature: 0.7 |
|
min_length: 30 |
|
max_length: 120 |
|
|
|
--- |
|
|
|
<!-- This model card has been generated automatically according to the information the Trainer had access to. You |
|
should probably proofread and complete it, then remove this comment. --> |
|
|
|
# flan-t5-base-xsum-swe |
|
|
|
This model is a fine-tuned version of [google/flan-t5-base](https://huggingface.co/google/flan-t5-base) on the None dataset. |
|
It achieves the following results on the evaluation set: |
|
- Loss: 1.4174 |
|
- Rouge1: 20.3004 |
|
- Rouge2: 6.2309 |
|
- Rougel: 17.5863 |
|
- Rougelsum: 17.5827 |
|
- Gen Len: 18.9998 |
|
|
|
## Model description |
|
|
|
More information needed |
|
|
|
## Intended uses & limitations |
|
|
|
More information needed |
|
|
|
## Training and evaluation data |
|
|
|
More information needed |
|
|
|
## Training procedure |
|
|
|
### Training hyperparameters |
|
|
|
The following hyperparameters were used during training: |
|
- learning_rate: 5e-05 |
|
- train_batch_size: 16 |
|
- eval_batch_size: 16 |
|
- seed: 42 |
|
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 |
|
- lr_scheduler_type: linear |
|
- num_epochs: 5 |
|
|
|
### Training results |
|
|
|
| Training Loss | Epoch | Step | Validation Loss | Rouge1 | Rouge2 | Rougel | Rougelsum | Gen Len | |
|
|:-------------:|:-----:|:-----:|:---------------:|:-------:|:------:|:-------:|:---------:|:-------:| |
|
| 1.8384 | 1.0 | 12751 | 1.6491 | 19.0409 | 5.1339 | 16.3642 | 16.3725 | 19.0 | |
|
| 1.6806 | 2.0 | 25502 | 1.5189 | 19.882 | 5.7451 | 17.1187 | 17.1233 | 19.0 | |
|
| 1.5961 | 3.0 | 38253 | 1.4575 | 20.237 | 6.1009 | 17.4642 | 17.4647 | 18.9998 | |
|
| 1.5556 | 4.0 | 51004 | 1.4268 | 20.2998 | 6.2217 | 17.5423 | 17.541 | 18.9998 | |
|
| 1.5429 | 5.0 | 63755 | 1.4174 | 20.3004 | 6.2309 | 17.5863 | 17.5827 | 18.9998 | |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.25.1 |
|
- Pytorch 1.13.0+cu116 |
|
- Datasets 2.8.0 |
|
- Tokenizers 0.13.2 |
|
|
