library_name: tf-keras | |
tags: | |
- sentence-similarity | |
## Model description | |
This repo contains the model and the notebook for fine-tuning BERT model on SNLI Corpus for Semantic Similarity. [Semantic Similarity with BERT](https://keras.io/examples/nlp/semantic_similarity_with_bert/). | |
Full credits go to [Mohamad Merchant](https://twitter.com/mohmadmerchant1) | |
Reproduced by [Vu Minh Chien](https://www.linkedin.com/in/vumichien/) | |
Motivation: Semantic Similarity determines how similar two sentences are, in terms of their meaning. In this tutorial, we can fine-tune BERT model and use it to predict the similarity score for two sentences. | |
## Training and evaluation data | |
This example demonstrates the use of the Stanford Natural Language Inference (SNLI) Corpus to predict semantic sentence similarity with Transformers. | |
- Total train samples: 100000 | |
- Total validation samples: 10000 | |
- Total test samples: 10000 | |
Here are the "similarity" label values in SNLI dataset: | |
- Contradiction: The sentences share no similarity. | |
- Entailment: The sentences have a similar meaning. | |
- Neutral: The sentences are neutral. | |
## Training procedure | |
### Training hyperparameters | |
The following hyperparameters were used during training: | |
| Hyperparameters | Value | | |
| :-- | :-- | | |
| name | Adam | | |
| learning_rate | 9.999999747378752e-06 | | |
| decay | 0.0 | | |
| beta_1 | 0.8999999761581421 | | |
| beta_2 | 0.9990000128746033 | | |
| epsilon | 1e-07 | | |
| amsgrad | False | | |
| training_precision | float32 | | |
## Model Plot | |
<details> | |
<summary>View Model Plot</summary> | |
![Model Image](./model.png) | |
</details> |