|
--- |
|
license: bigscience-bloom-rail-1.0 |
|
language: |
|
- fr |
|
- en |
|
pipeline_tag: text-classification |
|
base_model: |
|
- cmarkea/bloomz-3b-sft-chat |
|
--- |
|
|
|
|
|
Bloomz-3b-guardrail |
|
--------------------- |
|
|
|
We introduce the Bloomz-3b-guardrail model, which is a fine-tuning of the [Bloomz-3b-sft-chat](https://huggingface.co/cmarkea/bloomz-3b-sft-chat) model. This model is designed to detect the toxicity of a text in five modes: |
|
|
|
* Obscene: Content that is offensive, indecent, or morally inappropriate, especially in relation to social norms or standards of decency. |
|
* Sexual explicit: Content that presents explicit sexual aspects in a clear and detailed manner. |
|
* Identity attack: Content that aims to attack, denigrate, or harass someone based on their identity, especially related to characteristics such as race, gender, sexual orientation, religion, ethnic origin, or other personal aspects. |
|
* Insult: Offensive, disrespectful, or hurtful content used to attack or denigrate a person. |
|
* Threat: Content that presents a direct threat to an individual. |
|
|
|
This kind of modeling can be ideal for monitoring and controlling the output of generative models, as well as measuring the generated degree of toxicity. |
|
|
|
Training |
|
-------- |
|
|
|
The training dataset consists of 500k examples of comments in English and 500k comments in French (translated by Google Translate), each annotated with a probablity toxicity severity. The dataset used is provided by [Jigsaw](https://jigsaw.google.com/approach/) as part of a Kaggle competition : [Jigsaw Unintended Bias in Toxicity Classification](https://www.kaggle.com/competitions/jigsaw-unintended-bias-in-toxicity-classification/data). As the score represents the probability of a toxicity mode, an optimization goal of cross-entropy type has been chosen: |
|
$$loss=l_{\mathrm{obscene}}+l_{\mathrm{sexual\_explicit}}+l_{\mathrm{identity\_attack}}+l_{\mathrm{insult}}+l_{\mathrm{threat}}$$ |
|
with |
|
$$l_i=\frac{-1}{\vert\mathcal{O}\vert}\sum_{o\in\mathcal{O}}\mathrm{score}_{i,o}\log(\sigma(\mathrm{logit}_{i,o}))+(\mathrm{score}_{i,o}-1)\log(1-\sigma(\mathrm{logit}_{i,o}))$$ |
|
Where sigma is the sigmoid function and O represents the set of learning observations. |
|
|
|
Benchmark |
|
--------- |
|
|
|
Pearson's inter-correlation was chosen as a measure. Pearson's inter-correlation is a measure ranging from -1 to 1, where 0 represents no correlation, -1 represents perfect negative correlation, and 1 represents perfect positive correlation. The goal is to quantitatively measure the correlation between the model's scores and the scores assigned by judges for 730 comments not seen during training. |
|
|
|
| Model | Language | Obsecene (x100) | Sexual explicit (x100) | Identity attack (x100) | Insult (x100) | Threat (x100) | Mean | |
|
|------------------------------------------------------------------------------:|:---------|:-----------------------:|:-----------------------------:|:-----------------------------:|:--------------------:|:--------------------:|:----:| |
|
| [Bloomz-560m-guardrail](https://huggingface.co/cmarkea/bloomz-560m-guardrail) | French | 64 | 74 | 72 | 70 | 58 | 68 | |
|
| [Bloomz-560m-guardrail](https://huggingface.co/cmarkea/bloomz-560m-guardrail) | English | 63 | 63 | 62 | 70 | 51 | 62 | |
|
| [Bloomz-3b-guardrail](https://huggingface.co/cmarkea/bloomz-3b-guardrail) | French | 71 | 82 | 84 | 77 | 77 | 78 | |
|
| [Bloomz-3b-guardrail](https://huggingface.co/cmarkea/bloomz-3b-guardrail) | English | 74 | 76 | 79 | 76 | 79 | 77 | |
|
|
|
With a correlation of approximately 65 for the 560m model and approximately 80 for the 3b model, the output is highly correlated with the judges' scores. |
|
|
|
|
|
Opting for the maximum of different modes results in a score extremely close to the target toxicity of the original dataset, with a correlation of 0.976 and a mean absolute error of 0.013±0.04. Therefore, this approach serves as a robust approximation for evaluating the overall performance of the model, transcending rare toxicity modes. Taking a toxicity threshold ≥ 0.5 to create the target, we have 240 positive cases out of 730 observations. Consequently, we will determine the Precision-Recall AUC, ROC AUC, accuracy, and the F1-score. |
|
|
|
| Model | Language | PR AUC (%) | ROC AUC (%) | Accuracy (%) | F1-score (%) | |
|
|------------------------------------------------------------------------------:|:---------|:-------------:|:-----------------:|:------------------:|:---------------:| |
|
| [Bloomz-560m-guardrail](https://huggingface.co/cmarkea/bloomz-560m-guardrail) | French | 77 | 85 | 78 | 60 | |
|
| [Bloomz-560m-guardrail](https://huggingface.co/cmarkea/bloomz-560m-guardrail) | English | 77 | 84 | 79 | 62 | |
|
| [Bloomz-3b-guardrail](https://huggingface.co/cmarkea/bloomz-3b-guardrail) | French | 82 | 89 | 84 | 72 | |
|
| [Bloomz-3b-guardrail](https://huggingface.co/cmarkea/bloomz-3b-guardrail) | English | 80 | 88 | 82 | 70 | |
|
|
|
How to Use Bloomz-3b-guardrail |
|
-------------------------------- |
|
|
|
The following example utilizes the API Pipeline of the Transformers library. |
|
|
|
```python |
|
from transformers import pipeline |
|
|
|
guardrail = pipeline("text-classification", "cmarkea/bloomz-3b-guardrail") |
|
|
|
list_text: List[str] = [...] |
|
result = guardrail( |
|
list_text, |
|
return_all_scores=True, # Crucial for assessing all modalities of toxicity! |
|
function_to_apply='sigmoid' # To ensure obtaining a score between 0 and 1! |
|
) |
|
``` |
|
|
|
Citation |
|
-------- |
|
|
|
```bibtex |
|
@online{DeBloomzGuard, |
|
AUTHOR = {Cyrile Delestre}, |
|
ORGANIZATION = {Cr{\'e}dit Mutuel Ark{\'e}a}, |
|
URL = {https://huggingface.co/cmarkea/bloomz-3b-guardrail}, |
|
YEAR = {2023}, |
|
KEYWORDS = {NLP ; Transformers ; LLM ; Bloomz}, |
|
} |
|
``` |
