| --- |
| language: |
| - en |
| license: apache-2.0 |
| datasets: |
| - glue |
| metrics: |
| - pearsonr |
| model-index: |
| - name: t5-base-finetuned-stsb |
| results: |
| - task: |
| name: Text Classification |
| type: text-classification |
| dataset: |
| name: GLUE STS-B |
| type: glue |
| args: stsb |
| metrics: |
| - name: Pearson Correlation |
| type: pearson_correlation |
| value: 0.8937 |
| --- |
| |
|
|
| # T5-base-finetuned-stsb |
|
|
| <!-- Provide a quick summary of what the model is/does. --> |
|
|
| This model is T5 fine-tuned on GLUE STS-B dataset. It acheives the following results on the validation set |
| - Pearson Correlation Coefficient: 0.8937 |
|
|
|
|
| ## Model Details |
| T5 is an encoder-decoder model pre-trained on a multi-task mixture of unsupervised and supervised tasks and for which each task is converted into a text-to-text format. |
|
|
| ## Training procedure |
|
|
| ### Tokenization |
| Since, T5 is a text-to-text model, the labels of the dataset are converted as follows: |
| For each example, a sentence as been formed as **"stsb sentence1: " + stsb_sent1 + "sentence2: " + stsb_sent2** and fed to the tokenizer to get the **input_ids** and **attention_mask**. |
| Unlike other **GLUE** tasks, STS-B is a regression task where the goal is to predict a similarity score between 1 and 5. I have used the same stratey as descibed in the T5 paper for fine-tuning. In the paper, it is mentioned as |
| ``` We found that most of these scores were annotated in increments of 0.2, so we simply rounded any score to the nearest increment of 0.2 and converted the result to a literal string representation of the number (e.g. the floating-point value 2.57 would be mapped to the string “2.6”). At test time, if the model outputs a string corresponding to a number between 1 and 5, we convert it to a floating-point value; otherwise, we treat the model’s prediction as incorrect. This effectively recasts the STS-B |
| regression problem as a 21-class classification problem. ``` |
|
|
|
|
| ### Training hyperparameters |
|
|
| The following hyperparameters were used during training: |
| - learning_rate: 3e-4 |
| - train_batch_size: 32 |
| - eval_batch_size: 32 |
| - seed: 42 |
| - optimizer: epsilon=1e-08 |
| - num_epochs: 3.0 |
|
|
| ### Training results |
|
|
|
|
| |Epoch | Training Loss | Validation Pearson Correlation Coefficient | |
| |:----:|:-------------:|:-------------------:| |
| | 1 | 0.8623 | 0.8200 | |
| | 2 | 0.7782 | 0.8675 | |
| | 3 | 0.7040 | **0.8937** | |
