---
license: apache-2.0
tags:
- generated_from_keras_callback
model-index:
- name: tmp3y468_8j
  results: []
widget:
- text: "Ich liebe dich"
  example_title: "Non-vulgar"
- text: "Ich hasse dich"
  example_title: "Vulgar"
---

# KIZervus

This model is a fine-tuned version of [distilbert-base-german-cased](https://huggingface.co/distilbert-base-german-cased).
It is trained to classify german text into the classes "vulgar" speech and "non-vulgar" speech. 
The data set is a collection of other labeled sources in german. For an overview, see the github repository here: 
It achieves the following results on the evaluation set:
- Train Loss: 0.4221
- Train Accuracy: 0.8025
- Validation Loss: 0.4418
- Validation Accuracy: 0.8094
- Epoch: 2

## Training procedure

### Training hyperparameters

The following hyperparameters were used during training:
- optimizer: {'name': 'Adam', 'learning_rate': {'class_name': 'PolynomialDecay', 'config': {'initial_learning_rate': 5e-05, 'decay_steps': 1233, 'end_learning_rate': 0.0, 'power': 1.0, 'cycle': False, 'name': None}}, 'decay': 0.0, 'beta_1': 0.9, 'beta_2': 0.999, 'epsilon': 1e-08, 'amsgrad': False}
- training_precision: float32

### Training results

| Train Loss | Train Accuracy | Validation Loss | Validation Accuracy | Epoch |
|:----------:|:--------------:|:---------------:|:-------------------:|:-----:|
| 0.4524     | 0.7813         | 0.4397          | 0.7969              | 0     |
| 0.4215     | 0.8030         | 0.4838          | 0.7781              | 1     |
| 0.4221     | 0.8025         | 0.4418          | 0.8094              | 2     |


### Framework versions

- Transformers 4.21.1
- TensorFlow 2.8.2
- Datasets 2.2.2
- Tokenizers 0.12.1

### Supporter

![BMBF Logo](./BMBF_Logo.png)