File size: 4,625 Bytes
4d49759 9833d6c 4d49759 9833d6c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 |
---
license: apache-2.0
base_model: distilroberta-base
tags:
- generated_from_trainer
- rejection
- no_answer
- chatgpt
metrics:
- accuracy
- recall
- precision
- f1
model-index:
- name: distilroberta-base-rejection-v1
results: []
language:
- en
pipeline_tag: text-classification
co2_eq_emissions:
emissions: 0.07987621556153969
source: code carbon
training_type: fine-tuning
datasets:
- argilla/notus-uf-dpo-closest-rejected
---
# Model Card for distilroberta-base-rejection-v1
This model is a fine-tuned version of [distilroberta-base](https://huggingface.co/distilroberta-base) on multiple combined datasets of rejections from different LLMs and normal responses from RLHF datasets.
It aims to identify rejections in LLMs when the prompt doesn't pass content moderation, classifying inputs into two categories: `0` for normal outputs and `1` for rejection detected.
It achieves the following results on the evaluation set:
- Loss: 0.0544
- Accuracy: 0.9887
- Recall: 0.9810
- Precision: 0.9279
- F1: 0.9537
## Model details
- **Fine-tuned by:** ProtectAI.com
- **Model type:** distilroberta-base
- **Language(s) (NLP):** English
- **License:** Apache license 2.0
- **Finetuned from model:** [distilroberta-base](https://huggingface.co/distilroberta-base)
## Intended Uses & Limitations
It aims to identify rejection, classifying inputs into two categories: `0` for normal output and `1` for rejection detected.
The model's performance is dependent on the nature and quality of the training data. It might not perform well on text styles or topics not represented in the training set.
Additionally, `distilroberta-base` is case-sensitive model.
## How to Get Started with the Model
### Transformers
```python
from transformers import AutoTokenizer, AutoModelForSequenceClassification, pipeline
import torch
tokenizer = AutoTokenizer.from_pretrained("ProtectAI/distilroberta-base-rejection-v1")
model = AutoModelForSequenceClassification.from_pretrained("ProtectAI/distilroberta-base-rejection-v1")
classifier = pipeline(
"text-classification",
model=model,
tokenizer=tokenizer,
truncation=True,
max_length=512,
device=torch.device("cuda" if torch.cuda.is_available() else "cpu"),
)
print(classifier("Sorry, but I can't assist with that."))
```
### Optimum with ONNX
Loading the model requires the [🤗 Optimum](https://huggingface.co/docs/optimum/index) library installed.
```python
from optimum.onnxruntime import ORTModelForSequenceClassification
from transformers import AutoTokenizer, pipeline
tokenizer = AutoTokenizer.from_pretrained("ProtectAI/distilroberta-base-rejection-v1", subfolder="onnx")
model = ORTModelForSequenceClassification.from_pretrained("ProtectAI/distilroberta-base-rejection-v1", export=False, subfolder="onnx")
classifier = pipeline(
task="text-classification",
model=model,
tokenizer=tokenizer,
truncation=True,
max_length=512,
)
print(classifier("Sorry, but I can't assist with that."))
```
## Training and evaluation data
The model was trained on a custom dataset from multiple open-source ones. We used ~10% rejections and ~90% of normal outputs.
We used the following papers when preparing the datasets:
- [Do-Not-Answer: A Dataset for Evaluating Safeguards in LLMs](https://arxiv.org/abs/2308.13387)
- [I'm Afraid I Can't Do That: Predicting Prompt Refusal in Black-Box Generative Language Models](https://arxiv.org/abs/2306.03423)
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 3
### Training results
| Training Loss | Epoch | Step | Validation Loss | Accuracy | Recall | Precision | F1 |
|:-------------:|:-----:|:-----:|:---------------:|:--------:|:------:|:---------:|:------:|
| 0.0525 | 1.0 | 3536 | 0.0355 | 0.9912 | 0.9583 | 0.9675 | 0.9629 |
| 0.0219 | 2.0 | 7072 | 0.0312 | 0.9919 | 0.9917 | 0.9434 | 0.9669 |
| 0.0121 | 3.0 | 10608 | 0.0350 | 0.9939 | 0.9905 | 0.9596 | 0.9748 |
### Framework versions
- Transformers 4.36.2
- Pytorch 2.1.2+cu121
- Datasets 2.16.1
- Tokenizers 0.15.0
```
@misc{distilroberta-base-rejection-v1,
author = {ProtectAI.com},
title = {Fine-Tuned DistilRoberta-Base for Rejection in the output Detection},
year = {2024},
publisher = {HuggingFace},
url = {https://huggingface.co/ProtectAI/distilroberta-base-rejection-v1},
}
``` |